Trends in Patent Applications for Technologies in the Automotive Industry: Applications of Deep Learning and Machine Learning

Abstract

This study investigates global innovation trends in machine learning (ML) and deep learning (DL) technologies within the automotive sector through a patent analysis of 5314 applications filed between 2005 and 2022 across the five major patent offices (IP5). Using Cooperative Patent Classification (CPC) codes and keyword analysis, we identify seven sub-technology domains and examine both geographical and corporate patenting strategies. Our findings show that the United States dominates in overall filings, while Japan demonstrates a notably high share of triadic patents, which reflects a strong global-reach strategy. Patent activity is heavily concentrated in vehicle control and infrastructure traffic control, with emerging growth observed in battery management and occupant analytics. In contrast, security-related technologies remain underrepresented, indicating a potential blind spot in current innovation efforts. Corporate strategies diverge markedly; for example, some firms, such as Toyota and Bosch, pursue balanced tri-regional protection, whereas others, including Ford and GM, focus on dual-market coverage in the United States and China. These patterns illustrate how market priorities, regulatory environments, and technological objectives influence patenting behavior. By mapping the technological and strategic landscape of ML/DL innovation in the automotive industry, this study provides actionable insights for industry practitioners seeking to optimize intellectual property portfolios and for policymakers aiming to address gaps such as automotive cybersecurity in future R&D agendas.

1. Introduction

In the past decade, the automotive industry has experienced a paradigm shift, driven by the rapid integration of artificial intelligence (AI) technologies, particularly machine learning (ML) and deep learning (DL). These technologies fundamentally redefine how vehicles perceive their environment, make decisions, and interact with the infrastructure, paving the way for fully autonomous and intelligent mobility systems. Numerous studies have emphasized this transformation, highlighting AI’s role in reshaping vehicular automation [1], the evolution of autonomous driving technologies [2], and the growing significance of AI-related standard essential patents [3]. Additionally, recent work has explored the implications of these developments for future industry landscapes [4] and examined the technological underpinnings of AI-powered mobility [5]. As the automotive sector evolves into a high-tech domain that intersects with information and communication technologies (ICT), analyzing patent data offers critical insights into both the direction and intensity of global innovation. In particular, patent analysis provides a systematic framework for understanding which countries, companies, and technological areas are leading in the race toward autonomous, connected, and sustainable transportation solutions.

One of the most transformative applications of ML/DL in this domain is in the development of autonomous driving systems (ADS). Through advances in convolutional neural networks (CNN), reinforcement learning, and sensor fusion, modern vehicles are increasingly capable of interpreting complex driving environments and making real-time decisions with minimal human intervention. These advancements have contributed to a significant growth in patent activity related to perception algorithms, sensor integration, object recognition, trajectory planning, and adaptive control systems. Prior studies have highlighted this trend by examining the role of deep learning in autonomous vehicle control and perception [6], the integration of learning-based modules into driving systems [7], and the broader application of deep learning techniques in real-world driving environments [8]. In particular, Grigorescu et al. [9] provide a comprehensive overview of deep learning architectures, including CNNs and DRL, in interpreting complex environments and making decisions for autonomous driving. Kuutti et al. [10] focus on deep learning-enhanced control systems, highlighting adaptive decision-making in real time. Khanum et al. [11] further explore DL-based motion planning frameworks such as lane assist and emergency braking. Moreover, recent developments have extended DRL to multi-objective optimization for automotive component design, such as turbine blades, showcasing its practical utility in real-world engineering problems [12].

More specifically, between 2000 and 2023, the global patent landscape has shown a pronounced emphasis on AI-based perception systems, vehicle-to-everything (V2X) communication, and advanced driver-assistance systems (ADAS) [13,14]. Importantly, the United States, Japan, the Republic of Korea, and the European Union have become recognized as key regions where patent activity has intensified, often supported by large-scale public–private R&D initiatives and policy support for smart mobility. In the European context, for example, patent filings often reflect efforts to converge AI and IoT technologies to reduce traffic fatalities and environmental impact [15,16]. China has shown strong leadership in the area of sustainable mobility and eco-driving, with studies highlighting its bibliometric research on electric vehicle development [17], the rapid evolution of new energy transportation systems [18], and national strategies that drive global competitiveness in this domain [19].

In addition to autonomous navigation, ML and DL technologies are increasingly being leveraged in other strategic automotive domains. These include object detection and obstacle avoidance systems, traffic signal prediction, route optimization, driver monitoring, in-vehicle infotainment personalization, and anomaly detection for predictive maintenance. The emergence of end-to-end DL models and transformer-based architectures has further widened the scope of ML deployment in real-time systems. Since 2016, patent filings in such domains have risen sharply, particularly in technologies related to pedestrian and cyclist detection, multi-modal sensor fusion, and policy learning for motion control [20,21]. Khan et al. emphasized the role of DL in safety-related applications such as pedestrian and vehicle detection, highlighting future challenges in real-world autonomous driving [22]. Similarly, Devi et al. surveyed the role of ML and DL in object detection and traffic signal recognition, which are essential for vehicle decision-making systems [23]. Ansari and R applied deep reinforcement learning to urban navigation scenarios, addressing object detection and obstacle avoidance through sensor fusion [24]. In addition, Santra et al. discussed recent progress in multimodal sensor fusion and learning algorithms for automotive perception tasks, including driver monitoring and collision avoidance [25]. These innovations underscore the growing reliance on data-driven models in achieving safe and efficient vehicle autonomy.

ML is also converging with complementary technologies such as blockchain, quantum computing, and 5G-based IoT infrastructure. This convergence enables novel use cases such as secure vehicle-to-cloud communication, decentralized identity and payment systems, and ultra-low-latency communication for collision avoidance. Patent data indicate increasing activity at the intersection of ML and cybersecurity. For instance, recent reviews have explored the integration of blockchain, IoT, and ML [26], as well as their combined role in improving network security [27]. In the vehicular context, surveys and frameworks have examined how ML and blockchain can secure intelligent transportation systems [28] and address cybersecurity challenges in connected vehicles [29]. Although promising, this remains a relatively underdeveloped area compared to perception and control systems.

Furthermore, collaborative innovation ecosystems involving both traditional automakers and ICT companies have resulted in co-patenting and cross-licensing strategies, especially in domains such as connected vehicle platforms and edge-AI integration [30,31]. Biswas and Wang [32] further illustrate that integrating IoT, edge intelligence, 5G, and blockchain in autonomous vehicle architectures inherently promotes co-patenting and licensing within such ecosystems. Vermesan et al. [33] discuss the emergence of electric, connected, autonomous, and shared (ECAS) vehicles, emphasizing how cooperation between automakers and ICT firms fosters intellectual property sharing. Musa et al. [34] examine the convergence of information-centric networks and edge intelligence in the Internet of Vehicles, noting that stakeholder collaboration supports innovative connected vehicle solutions. Bathla et al. [35] emphasize that AI–IoT partnerships among traditional and tech firms are essential for achieving advanced autonomous systems, often accompanied by co-patenting and cross-licensing agreements.

From a geographical perspective, distinct national patterns have emerged in patent strategies. U.S.-based entities often pursue broad protection through global filings, particularly in the United States Patent and Trademark Office (USPTO), the European Patent Office (EPO), and under the Patent Cooperation Treaty (PCT) framework. Japanese firms, especially Toyota and Honda, exhibit a balanced tri-regional patenting approach, with significant portfolios in Japan, the U.S., and China. Korean and Chinese applicants tend to focus on their domestic markets while selectively targeting high-priority export destinations such as the US and Europe [36,37]. These variations reflect both differences in industrial policy and strategic intent, ranging from domestic market consolidation to global leadership in emerging standards.

At the same time, patent analysis has proven to be a powerful method to anticipate future technological trajectories and identify innovation gaps. For example, Mejía and Kajikawa [38] highlight recent trends in patent analytics and their academic significance. Maghsoudi et al. [39] applied text mining and network analysis to forecast IoT trends, while Yu and Pan [40] used citation networks to trace blockchain technology evolution. Similarly, Srivastava and Jain [41] conducted a scoping review on the application of patent analysis in technology management. Techniques such as dynamic network analysis, citation clustering, and semantic embedding models have been employed to forecast dominant technology clusters and detect early-stage innovation. By systematically mapping patents across time, region, and technical category, researchers and policymakers can identify areas of technological convergence, white space opportunities, and emerging competitive threats. In this sense, patent analytics serve not only as a descriptive tool but also as a predictive one.

Against this backdrop, this study aims to analyze global patent trends in ML and DL technologies as applied to the automotive industry. By examining more than 5000 patent records filed between 2005 and 2025 across the five major patent offices (IP5: USPTO, EPO, KIPO, JPO and CNIPA), we seek to understand (i) which countries and companies are leading in ML-based automotive innovation, (ii) how national strategies differ in their intensity and internationalization of patenting, and (iii) which technological subfields—such as vehicle control, infrastructure traffic management, and cybersecurity—are emerging as hotspots of innovation. Despite a growing number of studies on AI applications in mobility, few have offered a systematic and comparative analysis of ML/DL-related patent activity across both geographical jurisdictions and technological subdomains. This study addresses this gap by providing an integrated landscape analysis of ML/DL innovation in the automotive sector. Accordingly, the primary research question that guides this investigation is, "How do national and corporate actors differ in their ML/DL patenting strategies for automotive applications, and what are the emerging sub-technological areas of concentration?" This study contributes to the literature by offering a comprehensive and data-driven mapping of the evolving automotive ML/DL landscape and provides information relevant to industrial R&D planning, IP strategy development, and international technology policy coordination. The remainder of this paper is structured as follows: Section 2 describes the data collection and classification methodology. Section 3 and Section 4 presents the empirical findings on global trends, national and corporate patenting strategies, and sub-technology distributions. Section 5 discusses key implications and limitations. Finally, Section 6 concludes the paper and outlines directions for future research.

2. Materials and Methods

This section details the data source, selection criteria, pre-processing steps, and analytical procedures employed to examine machine learning (ML) and deep learning (DL) patent activity in the global automotive sector. A schematic overview of the workflow is presented in Figure 1.

2.1. Data Source and Collection

Patent data were retrieved from TechDNA^®, a commercial analytics platform operated by a Korean provider. The dataset comprises all published or granted applications filed at the five largest patent authorities—USPTO (United States), CNIPA (China), EPO (Europe), JPO (Japan), and KIPO (Republic of Korea)—collectively referred to as IP5. The collection window spans from 1 January 2005 to 16 February 2025. Each record includes bibliographic metadata, Cooperative Patent Classification (CPC) codes, abstracts, and full-text claims.

To ensure comprehensiveness and reliability, TechDNA collects raw patent data through dedicated harvesting servers that query open-access databases, including KIPO, USPTO, and Espacenet. Additionally, selected international patents undergo cross-verification with Google Patents to detect inconsistencies in bibliographic or legal status information. TechDNA also supports advanced filtering by CPC subclass, triadic patent detection, and keyword clustering, which enhance the granularity and consistency required for cross-national and cross-domain comparisons in this study.

2.2. Patent Selection Criteria

To identify relevant patent documents within the scope of machine learning and deep learning applications in the automotive sector, a two-tiered filtering strategy was employed. First, to capture the vehicle-related domain, only those patents containing keyword stems such as vehicle or automotive, and classified under the CPC subclass B60 (vehicles, in general), were retained. Second, to capture the ML/DL scope, patents were required to include keyword stems such as machine learning or deep learning, or to contain at least one CPC code within the G06N20/* subclass, which corresponds to machine learning models, architectures, and training methods. In practice, the dataset was first filtered using G06N20/00 to ensure that all patents were ML/DL-related. Co-assigned CPC codes from the B60 family were then used to determine the specific automotive application domain. This explains why G06N20/00 itself does not appear as a separate sub-technology category in the results table: all patents already contain this code by design.

This rule-based approach thus represents a hybrid strategy—combining keyword filtering with CPC code constraints—to secure both thematic precision and coverage. The method also mitigates risks of including patents with misleading terminology or missing CPC tags. For quality control, keyword stems were derived from a preliminary corpus analysis, and representative CPC assignments were cross-validated against sample patents.

While this rule-based filtering strategy ensures replicability and thematic relevance, it may not capture patents that employ non-standard or emerging terminology or that lack specific CPC tagging due to classification delay or ambiguity. To mitigate this, keyword stems were derived from preliminary corpus analysis, and the CPC classes were cross-validated using example patents. However, the approach remains inherently limited by its reliance on predefined terms. Future studies may enhance recall by incorporating iterative refinement, expert validation, and natural language processing techniques to capture latent semantic structures beyond explicit keywords or codes.

2.3. Temporal Scope and Data Pre-Processing

Because patent documents are typically published 18 months after filing, counts for the most recent filing years are incomplete. Quantitative analyses therefore focus on applications filed up to 31 December 2022. Filings from 2023 to 2025 are reported qualitatively but excluded from statistical trend calculations. Following temporal adjustment, data cleaning was conducted to ensure analytical accuracy. Duplicate family members and low-relevance documents (e.g., housekeeping continuations or re-issues) were removed through rule-based filtering based on title similarity and CPC mismatches. The final corpus comprises 5314 unique patents. Their distribution across patent offices is summarised in Figure 2; the USPTO accounts for 2337 filings (44.0%), followed by CNIPA (957) and KIPO (552).

2.4. Sub-Technology Classification and Validation

At a granular level, patents were assigned to seven sub-technology domains: Vehicle Control, Infrastructure Traffic Control, Navigation Devices, Occupant Analytics, Battery Management, Security and Anti-Abuse, and Business Models. The CPC-to-domain mapping is shown in

Table 1. Mapping of CPC codes to the seven sub-technology domains.

Sub-Technology Domain	Representative CPC Codes
Vehicle Control	G05D1/00, B60W60/00, B60W50/11, B60W40/00, B60W40/09
Infrastructure Traffic Control	G05D2201
Occupant Analytics	G06V20/55
Navigation Devices	G06N3/08, G06N3/04
Battery Management	H01M10/44, H01M10/48
Security and Anti-Abuse	H04L63/00, H04W12/06
Business Models	G06Q10/06, G06Q50/28

.

For patents associated with multiple CPC codes spanning different sub-technology domains, a de-duplication procedure was applied. In practice, the representative CPC was determined primarily based on the CPC assignment order in the KIPRIS database, which generally reflects the primary technical contribution. When the assignment order alone was insufficient to determine the most relevant category, the patent’s abstract and claims were briefly reviewed to confirm alignment with the pre-defined domain definitions. This procedure ensured that each patent was counted only once in the aggregated statistics.

To check the plausibility of the CPC-to-domain mapping, a random sample of 100 patents was reviewed by two domain experts from TechDNA^®, each with extensive experience in patent search and classification. Their independent assessments showed a high level of agreement on the assigned domains, supporting the reliability of the classification approach.

While this classification enables a detailed exploration of sub-technological trends, certain limitations must be acknowledged. Publication-lag adjustment may still under-represent late-stage filings in 2022, potentially leading to a slight underestimation of recent activity. In addition, the CPC-based classification framework can mislabel highly novel inventions that are not yet adequately represented in existing coding schemes. Finally, overlap between certain domains (e.g., vehicle control and infrastructure traffic control) cannot be completely eliminated due to shared technical scope. Despite these caveats, the resulting dataset provides a robust and comprehensive view of global ML/DL patenting activity in the automotive domain.

3. Global Trends in Machine Learning Patents for Automotive Applications

This section provides a comprehensive overview of global patenting activity in the field of machine learning (ML) and deep learning (DL) as applied to automotive technologies. By examining filing trends across major patent offices and analyzing the geographic and applicant-level distribution of patent portfolios, we aim to identify the strategic directions of key innovators and assess the international landscape of technological competition. The analysis draws from patent data filed between 2005 and early 2025, offering insights into both the growth trajectory and regional diversification of automotive ML/DL innovation efforts.

3.1. Global Filing Trends by Patent Office

Figure 2 illustrates the annual trends in patent applications related to machine learning (ML) and deep learning (DL) technologies applied to the automotive sector, based on filing data from six major patent authorities: the United States (USPTO), China (CNIPA), the Republic of Korea (KIPO), the European Patent Office (EPO), the World Intellectual Property Organization under the Patent Cooperation Treaty (PCT), and Japan (JPO). The data reveal a marked increase in patent filings beginning around 2015–2016, which coincides with significant industry investments in AI technologies by key players such as Toyota and Tesla, along with major advancements in deep learning algorithms. This inflection point represents a transition from exploratory development to active commercialization efforts across the automotive AI landscape.

The United States Patent and Trademark Office (USPTO) emerges as the dominant office in terms of total filings, accounting for 43.98% of all identified patents. The peak of U.S.-based filings occurred in 2020, with over 500 patent applications, reflecting an aggressive strategy to secure early-mover advantage in AI-driven vehicle technologies. China follows with a sharp increase starting in 2017 and peaking in 2020 with nearly 200 applications, suggesting an accelerated national R&D initiative in intelligent mobility. The Republic of Korea, Japan, and PCT filings also show consistent upward trends, although with relatively lower volumes. In particular, Korean applications experienced steady growth from 2015 onward, with a focus on vehicle control and battery optimization, while PCT applications, representing global patent strategies, also rose sharply during this period. The European Patent Office shows a gradual increase, consistent with its historically rigorous filing criteria.

The significant decline observed in the period post-2021 may be partially attributed to the inherent time lag between patent filing and publication, typically ranging from 18 to 24 months. As such, the apparent downturn in 2023–2025 should be interpreted cautiously, as many applications during this period may still be under review or pending publication. Overall, this trend analysis highlights the central role of the U.S. in leading innovation in AI-based automotive technologies, while also underscoring the increasing global participation of other regions such as China, the Republic of Korea, and Europe. The data suggest a high level of strategic activity and competition, with implications for both technological leadership and future market positioning in the autonomous and intelligent vehicle domain.

3.2. Cross-National Patent Filing Strategies

As shown in Figure 3, the United States leads ML/DL-related automotive patenting with 2048 IP5 filings (38.5% of the dataset) and 765 trilateral applications (37.3% of the total trilateral set). This dominance reflects its dual role as a major innovation hub and a primary commercial market for AI-enabled vehicle technologies.

Japan follows with 670 filings, of which 480 (71.6%) are trilateral—a strikingly high trilateral intensity that highlights a deliberate focus on securing IP rights in all major markets, especially the U.S. and Europe. Germany shows a similar profile, with 160 of its 262 filings (61.1%) being trilateral, underscoring an outward-oriented patent strategy.

In contrast, the Republic of Korea and China exhibit lower trilateral-to-IP5 ratios—26.3% and 43.8%, respectively. This suggests a prioritization of domestic or regional market protection, often followed by selective foreign expansion. In China’s case, local policy incentives and strong domestic demand appear to drive the emphasis on national filings, while the Republic of Korea’s approach reflects a focus on consolidating home-market leadership before broader internationalization.

These patterns point to two broad strategic models: (i) a global-reach strategy that emphasizes broad, simultaneous protection in multiple major jurisdictions (e.g., Japan and Germany), and (ii) a domestic-first strategy with staged international expansion (e.g., the Republic of Korea and China). While

Table 2. Number of ML/DL-related automotive patent applications by applicant nationality and patent type (IP5 and Trilateral Patents). Detailed figures illustrate the relative filing volumes and trilateral intensities across major countries, supporting the strategic classifications discussed in the text. Data source: Author’s analysis based on the TechDNA patent analytics platform (accessed February 2025).

Country	IP5	Trilateral Patent
USA	2048	765
Japan	670	480
Republic of Korea	585	154
China	322	141
Germany	262	160

summarizes the detailed counts, this interpretation highlights the strategic intent underlying the filing distributions rather than simply enumerating volumes.

3.3. Applicant-Level Patent Filing Strategies

Figure 4 and

Table 3. Patent application counts by company, segmented by patent office country code (KR, US, JP, EP, CN, WO). Data source: Author’s analysis based on TechDNA patent analytics platform (accessed February 2025).

Applicant	KR	US	JP	EP	CN	WO
TOYOTA MOTOR	7	78	74	15	58	1
FORD GLOBAL TECH	0	83	0	3	67	3
HYUNDAI	81	45	0	2	25	0
BOSCH	13	41	11	18	43	13
KIA	74	43	0	2	19	0
WAYMO	3	53	1	14	16	9
INTEL	4	32	6	20	22	9
GM GLOBAL TECH OPERATIONS	0	51	0	0	35	0
MOTIONAL AD	35	35	0	1	12	1
STATE FARM MUTUAL AUTOMOBILE INSURANCE	0	81	0	1	0	0

summarise the geographic composition of patent portfolios for the ten most active corporate applicants in ML/DL-enabled automotive technologies. The analysis reveals four dominant strategic patterns. The first pattern is a balanced tri-regional protection strategy, exemplified by Toyota and Bosch. These companies distribute filings relatively evenly across the United States, Japan (or Europe), and China. Toyota’s portfolio maintains a strong presence in each of these markets, while Bosch shows a similar diversification, reflecting its position as a global Tier-1 supplier embedded in multiple regional value chains. A second pattern, dual-market concentration, is followed by firms such as Ford and GM. This approach focuses on securing patents primarily in the United States and China, with minimal filings in Europe or Japan, suggesting an emphasis on defending core home markets while tapping into China’s rapidly growing mobility ecosystem. The third pattern, domestic market fortification with selective expansion, is pursued by Hyundai and Kia. Their portfolios are anchored by high domestic coverage through KIPO filings, while selectively targeting the United States and China for export-oriented protection. This reflects a phased globalization strategy that consolidates domestic leadership before committing to deeper overseas investment.

Finally, the niche or service-oriented targeting strategy is adopted by companies such as Waymo and State Farm. Waymo focuses filings in the United States, with targeted expansion into Europe and China to support autonomous taxi deployment. State Farm’s patents remain almost exclusively U.S.-based, consistent with its insurance-focused business model. By grouping applicants into these four categories, the narrative shifts from repetitive company-by-company enumeration toward a clearer comparative framework, while still providing sufficient detail to interpret the patterns shown in Figure 4 and Table 3.

4. Sub-Technology-Level Patent Analysis

To better understand the technological landscape of machine learning applications in the automotive sector, we classify and analyze patent filings across multiple sub-technology domains. This section provides a breakdown of overall patent volumes, investigates temporal trends in sub-technology development, and examines how key players distribute their patenting activity across these technical areas. Through this, we identify both dominant focus areas and emerging gaps in the innovation ecosystem.

4.1. Patent Trends by Key Sub-Technologies

To investigate how ML/DL technologies are applied across different functional areas within the automotive domain, we categorized all identified patents into key sub-technologies based on Cooperative Patent Classification (CPC) codes and relevant keywords. A total of 7 sub-technology domains were derived from the 5314 patents using a clustering approach applied to 96 frequently occurring CPC subclasses. These domains include Vehicle Control, Infrastructure Traffic Control, Navigation Devices, Occupant Analytics, Battery Management, Security and Anti-Abuse, and Business Models.

To identify technological focus areas in the domain of automotive machine learning, we conducted a comprehensive analysis of Cooperative Patent Classification (CPC) codes associated with the 5314 retrieved patents. Figure 5 and

Table 4. Number of applications by major CPC codes, presenting the ten most frequently occurring classifications in the ML/DL-based automotive patent corpus. Source: TechDNA patent analytics platform.

CPC Classification Code	Number of Applications
G06N20/00	4845
G06N3/08	1425
G05D1/00	1089
G06N3/04	762
G05D2201	724
B60W60/00	718
B60W50/11	620
B60W40/00	597
B60W40/09	563
G06V20/55	552

illustrate the top 10 most frequently cited CPC codes, offering insights into the dominant technical domains.

As expected, the most prevalent classification was G06N20/00, which corresponds to machine learning models or techniques, with 4845 occurrences. This category includes neural networks and deep learning systems, underscoring the central role of AI model development in automotive applications. The next most common classifications were G06N3/08 and G05D1/00, with 1425 and 1089 citations, respectively, corresponding to systems for adaptive control and general control of vehicular dynamics.

Several other CPCs reflected key functional domains of autonomous and intelligent driving. For example, B60W60/00 and B60W50/11 pertain to energy management and adaptive cruise control systems, while B60W40/00 and B60W40/09 deal with route planning and driver intent interpretation. The inclusion of G06V20/55, which relates to vision-based object detection, indicates the integration of perception systems into vehicular intelligence.

It is worth noting that many of these CPCs are associated with more than one patent, reflecting overlapping technological domains. This reinforces the multidisciplinary nature of AI-based automotive innovation, where navigation, control, perception, and optimization systems must interact seamlessly. Furthermore, the frequency distribution reveals a strong concentration on control and decision-making systems, suggesting that these remain the cornerstone of ongoing R&D efforts in the field.

The substantial representation of energy- and perception-related CPCs suggests that industry actors are increasingly converging toward integrated systems that manage driving behavior, optimize energy consumption, and ensure situational awareness. Overall, the CPC analysis offers a granular lens into the technical composition of machine learning innovations shaping the next generation of automotive technologies.

4.2. Temporal Trends in Sub-Technology Patent Filings

To better understand how technological focus has evolved over time within the automotive machine learning domain, Figure 6 presents the annual distribution of patent filings across seven key sub-technologies.

Table 5. Patent application trend by sub-technology from 2002 to February 2025, based on data retrieved from the TechDNA patent analytics platform.

Sub-Technology	Total Number of Patents
Vehicle Control	3109
Infrastructure Traffic Control	1699
Occupant Analytics	886
Navigation Devices	844
Battery Management	496
Security and Anti-Abuse	498
Business Models	275

summarizes the total number of patent applications in each category. This temporal analysis reveals both the growth trajectory and relative maturity of each subdomain. The data demonstrate a dramatic rise in patent activity beginning around 2015, reaching a peak between 2019 and 2021. Vehicle Control stands out as the most prominent category, with a total of 3109 filings. This dominance reflects the automotive industry’s intense focus on enabling fully or partially autonomous driving, particularly at SAE Level 4 and above. The significant growth in this category during the peak years suggests a period of accelerated R&D investment and prototyping in core vehicular intelligence systems.

Infrastructure Traffic Control follows with 1699 patents, showing steady growth that reflects the increasing importance of V2X (vehicle-to-everything) integration and smart city coordination. This category often overlaps with Navigation Devices (844 filings), suggesting a convergence of control and localization technologies required for real-time route optimization and traffic response. Occupant Analytics, with 886 filings, also exhibited notable momentum during the 2018–2021 period. This category includes innovations in driver monitoring systems, gesture recognition, and personalized in-cabin experiences, aligning with growing interest in human–machine interaction and safety compliance.

Battery Management (496 filings) and Security and Anti-Abuse (498 filings) have emerged more recently as critical enablers for electric and connected vehicles. The data show an upward trend in both areas post-2018, indicating rising concern over range optimization, energy safety, and cybersecurity threats such as sensor spoofing and data breaches. However, their overall volume remains modest compared to the leading categories, suggesting opportunities for further research and innovation. Lastly, Business Models accounted for 275 filings, reflecting the integration of machine learning into usage-based insurance, fleet management, and predictive maintenance services; while small in total share, this domain represents a shift toward data-driven services and value extraction. In addition to hardware functionality. In summary, Figure 6 and Table 5 together highlight how patenting behavior reflects both technological readiness and strategic industry shifts. Core control and infrastructure systems remain the mainstay of innovation, while emerging domains such as cybersecurity and service models are poised for growth in the coming years.

4.3. Applicant-Level Distribution of Sub-Technology Focus

To gain further insight into how leading global companies are strategically distributing their innovation efforts, Figure 7 and

Table 6. Patent count of major sub-technologies by applicant, based on cumulative filings from 2002 to February 2025. Data were obtained from the TechDNA patent analytics platform.

Applicant	Vehicle Control	Infrastructure Traffic Control	Occupant Analytics	Navigation Devices
TOYOTA	135	62	63	30
FORD	61	41	18	11
HYUNDAI	80	40	41	17
KIA	70	33	38	14
BOSCH	72	21	21	10
WAYMO	93	39	7	30
INTEL	70	21	32	12
ZOOXINC	70	36	7	19
GM	53	14	18	6
NOTIONAL AD	79	25	7	32

present a comparative analysis of patent filings by major applicants across four representative sub-technologies: Vehicle Control, Infrastructure Traffic Control, Occupant Analytics, and Navigation Devices. The bubble chart in Figure 7 visualizes the volume and focus of patent activity, while Table 6 summarizes absolute counts by firm and sub-domain. Vehicle Control emerges as the most dominant sub-technology for nearly all applicants. Toyota leads this category with 135 filings, followed by Waymo (93), Hyundai (80), and Notional AD (79). These figures highlight the centrality of advanced driver-assistance and autonomous driving technologies in the R&D portfolios of both traditional automakers and tech-driven mobility firms.

Infrastructure Traffic Control patents show a more distributed pattern. Toyota (62) again takes the lead, underscoring its dual focus on in-vehicle and V2X coordination systems. In particular, Ford (41), Hyundai (40), and Waymo (39) also report substantial activity, reflecting investment in connectivity infrastructure as a key enabler for real-time navigation and traffic optimization. Occupant Analytics—encompassing driver monitoring, gesture control, and cabin personalization—shows a more selective adoption pattern. Toyota again leads with 63 filings, closely followed by Hyundai (41), Kia (38), and Intel (32). These firms appear to prioritize in-cabin AI as a differentiator for user experience and safety, while companies like Waymo and Zoox, which emphasize full autonomy, show minimal investment in this area.

Navigation Devices present an interesting contrast. Toyota and Waymo each report 30 filings, suggesting parallel efforts in localization and route planning, albeit from different technological perspectives. Notional AD (32) and Zoox (19) also demonstrate strong positioning in this space, indicative of their focus on robotaxi deployment and urban mobility systems. Collectively, the data suggest that while all firms recognize the strategic value of vehicle control technologies, differences in emphasis across sub-domains reflect varying business models and deployment strategies. Traditional automakers such as Toyota and Hyundai exhibit balanced portfolios that span control, infrastructure, and user interaction. In contrast, companies like Waymo and Zoox show greater specialization, focusing on fully autonomous systems with tailored infrastructure support.

This multi-dimensional comparison reinforces the notion that sub-technology specialization is influenced not only by technological capability but also by target markets and organizational roles within the evolving mobility ecosystem.

5. Discussion

The results of this study provide a comprehensive view of the global innovation landscape for machine learning (ML) and deep learning (DL) technologies in the automotive sector. Several key insights emerge from the patent analysis that warrant further discussion, both in terms of technological focus and strategic behavior among leading applicants and jurisdictions.

First, the exponential growth in ML/DL-related patent filings from 2015 onward confirms that the automotive industry has entered a phase of accelerated technological transformation. This trend temporally aligns with notable advances in deep learning architectures and the proliferation of autonomous driving initiatives among OEMs and tech firms. The United States has gained prominence as the dominant jurisdiction in terms of both total filings and triadic patents, underscoring its central role in early innovation and intellectual property consolidation. However, the relatively high trilateral filing rates observed in Japan and Germany suggest a different strategy—one that prioritizes broad market protection in key global regions rather than sheer volume. In contrast, the lower trilateral intensity of the Republic of Korea and China may indicate a stronger focus on domestic deployment or phased internationalization strategies.

Second, our sub-technology analysis reveals that patent activity is heavily concentrated in Vehicle Control and Infrastructure Traffic Control, with these categories collectively accounting for more than 60% of total filings. This underscores a strategic emphasis on the core functionalities needed to achieve higher levels of driving autonomy. In particular, the high incidence of SAE Level 4-related filings suggests that firms are aggressively targeting use cases such as highway pilot systems, robotic taxis, and geofenced autonomous logistics. The growing presence of Battery Management and Occupant Analytics also indicates diversification, with patents increasingly addressing energy optimization and intelligent services centered on passengers. However, the relative underrepresentation of Security and Anti-Abuse technologies, despite their importance in ensuring system robustness and safety, warrants deeper exploration. Recent scholarly work suggests multiple contributing factors. First, strategic neglect may occur when OEMs prioritize visible, market-differentiating functionalities—such as advanced driver assistance or energy efficiency—over backend security features, which are often treated as compliance requirements rather than innovation drivers [42]. Second, the adoption of key standards, such as ISO/SAE 21434 [43] and UNECE WP.29 R155, [44] remains uneven across jurisdictions due to differing regulatory cultures and entrenched national regulatory frameworks [45]. This divergence can create institutional barriers that delay the integration of cybersecurity measures into core automotive engineering practices. Third, the inherent technical complexity of integrating robust security measures—ranging from diverse communication channels to secure software integration—poses a significant challenge for innovation [46]. These factors collectively contribute to the lower observed patenting activity in the security domain, despite its critical importance in ensuring the resilience and safety of connected and autonomous vehicles.

From a corporate strategy perspective, companies differ markedly in their geographic and technological focus, and these patterns are shaped by strategic and regulatory drivers. Toyota’s balanced tri-regional filing approach reflects its globally distributed R&D network and the need to secure IP in its highest-revenue markets. In contrast, Ford and GM’s concentration in North America and China mirrors both market prioritization and the competitive and regulatory environments in these jurisdictions. Hyundai’s and Kia’s more selective overseas filings may stem from a phased internationalization strategy shaped by resource allocation and domestic market consolidation before entry into regions with more stringent IP enforcement regimes. These interpretations suggest that filing geography is a deliberate outcome of intertwined economic, regulatory, and technological considerations rather than a byproduct of filing convenience.

It is interesting to note that the participation of non-traditional automotive actors such as Waymo, Intel, and State Farm highlights the expanding boundaries of automotive innovation, where software, semiconductor, and service-layer firms increasingly contribute to the technological ecosystem. From an industry perspective, these findings suggest that OEMs and tech companies are likely to continue converging on core domains such as perception, planning, and control, while simultaneously branching into areas like battery efficiency and user-centric features. This dual focus is expected to shape future vehicle architectures that are modular, software-defined, and energy-optimized. For scientific research, the results underline the importance of interdisciplinary collaboration, particularly in areas like secure AI, human–machine interaction, and scalable computing infrastructure. The patent landscape signals several nascent subfields—e.g., personalized occupant services or predictive vehicle diagnostics—that merit deeper investigation. From a consumer standpoint, the observed innovation patterns imply that upcoming vehicles will likely offer enhanced autonomy, personalization, and energy management capabilities. This creates new expectations around transparency, safety, and user trust, especially as AI-driven decisions become more embedded in everyday driving experiences.

Although this study offers a detailed and data-driven perspective, it is not without limitations. First, the analysis is based solely on patent documents, which may not capture the full spectrum of R&D activity—particularly in cases of trade secrets or unpublished applications. Second, although CPC codes provide a robust taxonomy for technical classification, they may not fully capture emerging hybrid technologies or cross-domain innovations. Third, due to the inherent publication delay of 18–24 months, the observed decline in patent filings in recent years (post-2022) may reflect an artifact of data availability rather than an actual downturn in innovation. Fourth, the keyword and CPC-based filtering method employed in this study, while systematic and reproducible, may have excluded patents that utilize unconventional terminology or novel classification schemes.

Future research should consider integrating additional data sources such as technical papers, funding trends, and product deployment data to triangulate the maturity and commercial readiness of ML/DL technologies in automotive contexts. In addition, qualitative studies involving interviews with R&D managers or IP strategists could enrich our understanding of the decision-making processes behind patent filing strategies. Sensitivity analyses of keyword and classification filters, as well as collaboration with domain experts, would also improve the reliability and inclusiveness of future patent-based investigations. Lastly, comparative analyses across other sectors, such as aviation or smart manufacturing, may help contextualize the uniqueness or commonality of trends observed in the automotive domain. Furthermore, longitudinal studies tracking the translation of patent filings into commercial products could offer valuable insights into the innovation-to-market pipeline, especially in fast-evolving subdomains such as autonomous logistics and AI-based driver assistance. In summary, this study highlights both the dynamism and asymmetry of global ML/DL innovation in the automotive field. As the industry moves toward increasing levels of autonomy and connectivity, understanding patent trends will be vital for policymakers, firms, and researchers alike in shaping competitive and regulatory strategies.

6. Conclusions

This study provides a comprehensive analysis of the technological and geographical dynamics of patent activity in the automotive machine learning (ML) and deep learning (DL) domain, based on 5314 patent applications filed in IP5 offices from 2005 to early 2025. The findings reveal different national strategies and corporate behaviors shaping the global innovation landscape for intelligent mobility. The United States emerges as the dominant jurisdiction, accounting for 44% of all filings, underscoring its central role in the development of AI-based automotive technologies. However, filing behaviors vary considerably between countries. Japanese firms, most importantly Toyota, show a triregional filing strategy, seeking balanced protection in the United States, Europe, and China. This contrasts with Korean and Chinese companies, which focus heavily on their home markets while selectively targeting key foreign jurisdictions such as the US and China. Particularly noteworthy is Japan’s high proportion of tri-polar filings (72% of its IP5 portfolio), indicating a deliberate and globally oriented patenting strategy aimed at securing competitive advantage across leading markets.

From a technological perspective, vehicle control dominates the patent landscape, comprising 55% of all applications. Among these, over 70% relate to advanced autonomy systems at SAE Level 4 or higher, reflecting global R&D priorities in achieving full self-driving capability. Furthermore, substantial growth in infrastructure traffic control (31%) and navigation technologies (18%) indicates increased emphasis on V2X coordination and integration of smart cities. In contrast, cybersecurity-related patents remain scarce, accounting for just 2.3% of the total corpus, which highlights a critical gap in safeguarding autonomous systems against hacking, spoofing, and other digital threats. The strong overlap between vehicle control and battery management technologies also suggests a growing interest in integrated software–hardware optimization for electric and autonomous vehicles.

Corporate strategies further underscore the heterogeneous nature of innovation within this domain; while firms such as Toyota and Bosch allocate substantial resources to battery management, others such as GM concentrate on occupant analytics. Hyundai and Kia exhibit a strong focus on domestic filings, particularly in vehicle control, which appears aligned with their home-market consolidation strategy. Meanwhile, emerging mobility firms such as Waymo and Motional pursue optimized targeted portfolios for the deployment of robotaxi services in select regions.

These findings collectively suggest a multipolar structure in the global automotive ML/DL ecosystem, where firms and nations compete not only through volume of filings but also through differentiated technological focus and regional positioning. For policymakers, we recommend the creation of targeted funding and regulatory support programs specifically for underdeveloped areas such as automotive cybersecurity. This could include public–private R&D grant schemes, international collaboration frameworks for security testing, and integration of cybersecurity readiness into vehicle type-approval processes. Clear performance indicators (e.g., number of certified secure vehicle platforms, reduction in reported cyber incidents) should be established to monitor progress.

For industry stakeholders, firms should align intellectual property strategies with both market expansion plans and anticipated regulatory requirements. This may involve forming cross-industry consortia to pool patents in strategic domains, conducting competitive patent landscaping every 12–18 months to identify white-space opportunities, and integrating IP analytics into product development roadmaps. The success of these measures can be assessed through metrics such as licensing revenue growth, coverage of strategic markets, and acceleration of time-to-market for AI-enabled vehicle features.

For academia, we encourage closer collaboration with industry and government agencies to bridge the gap between early-stage research and commercial deployment. This can be achieved by embedding doctoral and postdoctoral researchers within corporate R&D teams, co-developing open-source datasets and benchmarking tools for automotive AI and cybersecurity, and jointly publishing implementation guidelines. Impact can be measured by the adoption rate of academic research in commercial systems, participation in standards development committees, and increased industry-funded research projects.

However, this study is not without limitations. Patent data alone cannot fully capture trade secrets or R&D efforts that have yet to be published, and the time delay in publication introduces a degree of uncertainty in recent trends. Future research could benefit from incorporating complementary data sources such as scientific publications, funding records and product release information, as well as deeper analyses of individual firm strategies and the market implications of their patent holdings. In particular, longitudinal tracking of how patented technologies transition into commercial vehicle platforms would help assess the real-world impact of current innovation trends. In conclusion, this work provides an empirical foundation for understanding the technological evolution of ML/DL in the automotive industry and offers practical, stakeholder-specific guidance for shaping R&D and IP policy in a rapidly transforming mobility landscape.