Harmonized Autonomous–Human Vehicles via Simulation for Emissions Reduction in Riyadh City

Comment R1.1

This study evaluates the impact of different AV penetration scenarios (0%, 10%, 30%, 50%) on traffic performance and carbon emissions along Prince Mohammed bin Salman bin Abdulaziz Road in Riyadh, Saudi Arabia. The paper is well-structured and results are statistically significant. However, there are areas for improvement.

The authors need to discuss further the generalizability of findings to other similar cities (e.g., Dubai, Doha) and their policy implications (e.g., infrastructure upgrades or incentives).

Answer R1.1

We sincerely thank the reviewer for the encouraging feedback and for recognizing the structure and statistical rigor of our study. Your comment regarding the generalizability of the findings and the need to discuss broader policy implications is highly appreciated. In response, we have revised the Discussion section to reflect how the results may extend to other rapidly developing Gulf cities such as Dubai and Doha, which share similar urban dynamics and transportation challenges with Riyadh. We also elaborated on the practical policy implications of our work highlighting the importance of infrastructure preparedness for V2X communication, the potential benefits of phased AV integration strategies, and the role of government incentives in accelerating adoption. We believe these additions strengthen the relevance and impact of the study for both researchers and policymakers across the region.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is added as follows:

In addition to its local relevance, the outcomes of this study offer valuable insights for other rapidly developing Gulf cities such as Dubai and Doha, which share similar urban growth patterns, car-centric infrastructure, and ambitions for smart mobility transformation. The observed improvements in traffic flow and emission reductions through coordinated AV-HV integration are highly likely to be generalizable in these contexts. While specific traffic conditions may vary, the behavioral advantages of AV coordination such as reduced stops, smoother lane transitions, and lower emissions are broadly applicable. Furthermore, the study carries key policy implications, emphasizing the importance of infrastructure preparedness, such as enabling V2X-compatible intersections and deploying roadside units (RSUs). Policymakers may also consider incentive-based strategies, including subsidies for AV-enabled fleets and phased implementation along strategic corridors. Overall, the HARMONY-AV framework provides a flexible simulation approach to support data-driven mobility planning across the region.

Comment R1.2

More details on the cleaning process of TomTom and Saudi data need to be provided (e.g., outlier handling, timestamp alignment) and discuss data limitations (e.g., coverage of holidays or special events).

Answer R1.2

We appreciate the insightful suggestion to elaborate on the data preprocessing steps and acknowledge potential data limitations. In response, we have expanded Section 3.1 (Datasets and Sources) to provide more detailed explanations of the cleaning process for both TomTom and Saudi traffic datasets. Specifically, we now describe how outliers in traffic volume and travel time were identified using interquartile range (IQR) analysis and smoothed using local averaging techniques. Timestamp alignment across datasets was handled by converting all time data to a unified 24-hour format and correcting for minor inconsistencies in time zone metadata. Additionally, we have included a paragraph discussing limitations such as reduced data granularity during public holidays, school breaks, or major events, which may introduce some bias in traffic patterns. These clarifications strengthen the reproducibility and transparency of our data pipeline.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Datasets and Sources section: paragraph is as follows:

To evaluate the impact of AVs on traffic efficiency and emissions, a comprehensive dataset was constructed using both real-world and simulated sources. The raw datasets underwent standard preprocessing and cleaning procedures to ensure consistency and reliability. Redundant entries were removed, column names were normalized, and all timestamps were aligned to a consistent 24-hour format using UTC+3 to match the local time zone of Riyadh. Inconsistencies due to format differences between data providers were resolved through a parsing and validation script. Outlier handling was conducted using an interquartile range (IQR)-based method to detect and smooth anomalous traffic counts or travel times, especially those occurring due to sensor noise or atypical reporting. Detected outliers were corrected using a locally weighted smoothing function to preserve the general pattern without distorting peak or off-peak distributions. Intersections were geocoded and mapped to SUMO nodes using a Python-based matching algorithm that cross-referenced OpenStreetMap (OSM) data and official Saudi road maps. Vehicle types were proportionally allocated based on regional transport fleet distributions reported by the Saudi Ministry of Transport. While the dataset provides rich temporal coverage throughout the year, it does exhibit certain limitations. Traffic data collected during national holidays, religious events, and major public gatherings may not fully represent regular weekday patterns. These periods were flagged and considered in the analysis, though they were not removed to preserve overall traffic behavior variance. These datasets were ultimately selected for their accuracy, comprehensiveness, and relevance to the specific traffic conditions in Riyadh. An overview of the dataset sources is illustrated in Figure 2, and more details are discussed in the following subsections.

Comment R1.3

Please add a comparison between SUMO simulation results and field observations (if available) to validate model accuracy.

Answer R1.3

We sincerely thank the reviewer for this valuable suggestion regarding the validation of simulation results against field observations. While we acknowledge the importance of such comparative analysis, direct field observation data such as detailed vehicle trajectories, real-time emissions, or GPS-based trip profiles for the selected Riyadh corridor was not publicly available at the required granularity during the study period. As a result, a formal validation against field measurements could not be performed. However, we mitigated this limitation by anchoring our simulation inputs and assumptions to credible and localized data sources, including official traffic counts from Saudi open datasets and congestion profiles from the TomTom Traffic Index. Additionally, the simulation outputs such as travel times, stop frequencies, and speed profiles were found to be consistent with reported regional averages and published benchmarks for urban road segments in Riyadh. We have added a note in the Discussion section to reflect this limitation and clarify how input realism and statistical rigor contribute to the model’s credibility in the absence of field-based calibration.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is added as follows:

While direct field observation data for the selected corridor in Riyadh was not available at the necessary resolution, we grounded our simulation using official traffic counts from Saudi open data sources and congestion metrics from the TomTom Traffic Index. The resulting outputs such as average travel times and speed profiles aligned closely with published regional norms. Although formal calibration against real-time field measurements was not feasible, the use of validated datasets, combined with statistical testing, lends credibility to the findings and supports their relevance for practical urban mobility planning.

Comment R1.4

  The authors need to clearly explain the approximations used in SUMO for V2V/V2I/V2N communication (e.g., TraCI API call frequency) and their potential impact on results.

Answer R1.4

Thank you for pointing out the need for additional clarity regarding the communication approximations used in SUMO. We have now expanded Section 4 (Modeling AV Communication and Coordinated Behavior) to provide a more explicit description of how V2V, V2I, and V2N coordination was emulated using the TraCI API. Specifically, we clarified the call frequency of the TraCI API, which operated at a simulation step of 1 Hz (one API call per simulated second), enabling near-real-time interaction between vehicles and infrastructure components. We also discuss the implications of this approximation, noting that while the 1-second resolution captures general behavioral synchronization, it may smooth over finer-grained variations that could emerge with sub-second network latencies or denser AV deployments. These limitations have been explicitly stated in the Discussion section, reinforcing the need for future work using full-stack co-simulation environments like OMNeT++ or Eclipse MOSAIC to model packet loss, delays, and network congestion more accurately.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Modeling AV Communication and Coordinated Behavior section: paragraphs are added as follows:

The simulation of autonomous vehicle (AV) communication remains a challenge in microscopic traffic simulators like SUMO, which do not natively support explicit vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), or vehicle-to-network (V2N) protocols. However, the behavioral effects of these communication types can be approximated using parameter tuning, API interaction, and modular scripting. This section presents how such communication paradigms are represented in our simulations, as well as the assumptions and modules used to replicate real-world coordinated AV behavior.

AVs in this study were configured using the Krauss car-following model, which was adjusted to reflect coordinated behavior enabled by V2X systems. Parameters such as a reduced reaction time (τ = 0.5 s), shorter minimum following gap (minGap = 1.5 m), and enhanced acceleration/deceleration thresholds simulate smoother movement and anticipatory control. Cooperative lane-changing was enabled through lcStrategic and lcCooperative, allowing for predictive and synchronized merging, critical in congested intersections.

To emulate communication-driven coordination, we used the TraCI API to dynamically control vehicle behavior during simulation runs. The TraCI interface was called at a frequency of 1 Hz (one API call per simulated second), allowing each AV to retrieve traffic signal states, adjust speed relative to neighbors, and receive routing updates in near real time. Specifically, functions such as traci.vehicle.getNeighbors() and traci.trafficlight.getRedYellowGreenState() were employed to approximate V2V and V2I behavior respectively, while centralized logic governed V2N-like coordination. Although this 1-second resolution captures essential behavioral dynamics with manageable computational overhead, it abstracts finer network-layer effects such as sub-second communication delays, jitter, and packet loss. We acknowledge this limitation in the Discussion section and highlight the potential for future studies to incorporate full-stack V2X simulation platforms (e.g., OMNeT++ or Eclipse MOSAIC) for more granular validation and protocol-aware experimentation.

Comment R1.5

The actual deployment costs (e.g., sensors, communication infrastructure) and feasibility at 50% AV penetration need to be explored further.

Answer R1.5

Thank you for raising this important point regarding the real-world feasibility and associated costs of achieving high AV penetration rates, particularly at the 50% level. While our study is focused on simulation-based performance evaluation, we agree that acknowledging practical deployment constraints adds valuable context. To address this, we have added a paragraph to the Discussion section reflecting on the economic and infrastructural challenges of large-scale AV integration. We briefly discuss the capital costs of vehicle sensors (e.g., LiDAR, radar), the need for widespread deployment of roadside units (RSUs) and V2X-compatible traffic lights, and the supporting communication backbone (e.g., 5G or edge computing). While precise cost estimations were beyond the scope of this simulation-based work, we emphasize that our 50% scenario should be interpreted as a long-term planning target rather than a short-term policy recommendation. The updated section now also suggests that incremental deployment (e.g., along select high-priority corridors) may serve as a more feasible entry point for policymakers and stakeholders.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is added as follows:

It is also important to consider the real-world feasibility and cost implications of achieving high AV penetration rates, particularly the 50\% scenario used in this study. While our simulations illustrate the potential benefits of such integration, actual deployment would require substantial investment in AV hardware (e.g., LiDAR, radar, onboard processing units), as well as supporting infrastructure such as roadside units (RSUs), V2X-enabled traffic signals, and robust wireless communication networks (e.g., 5G or edge computing). The costs associated with sensorized fleets and continuous network maintenance may pose financial and logistical challenges, particularly in regions where existing infrastructure is not yet digitally mature. Therefore, our 50\% AV scenario should be viewed as a long-term planning benchmark rather than an immediate implementation goal. In practice, incremental adoption such as phased deployment along strategic corridors may offer a more feasible pathway to realizing the benefits of AV-human vehicle harmonization over time.

Comment R1.6

This paper needs to discuss the HARMONY-AV framework’s feasibility in Riyadh in more detail, including potential collaboration with local traffic authorities.

Answer R1.6

We appreciate the suggestion to discuss the practical feasibility of deploying the HARMONY-AV framework in Riyadh, including the potential for collaboration with local traffic authorities. In response, we have added a paragraph to the Discussion section elaborating on the framework's alignment with current smart city initiatives in Saudi Arabia and the role of institutional partnerships. Specifically, we highlight how the HARMONY-AV simulation approach could support planning within Riyadh’s digital transformation agenda and how collaboration with entities such as the Saudi Data and AI Authority (SDAIA) or the Royal Commission for Riyadh City could enhance data access, infrastructure integration, and policy development. These details help situate our work within a realistic roadmap toward deployment.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is added as follows:

Beyond simulation, the HARMONY-AV framework presents an opportunity for real-world application in Riyadh’s evolving smart mobility ecosystem. Saudi Arabia’s Vision 2030 and related initiatives led by entities such as the Royal Commission for Riyadh City and the Saudi Data and AI Authority (SDAIA)  are already fostering advancements in digital infrastructure and data-driven urban management. The framework developed in this study aligns well with these goals and could serve as a decision-support tool for evaluating AV integration strategies. Collaborations with local traffic authorities would be instrumental in refining the framework, enabling access to granular traffic data, pilot deployment zones, and validation platforms. Establishing such partnerships would also facilitate the gradual implementation of V2X infrastructure and policy development necessary for the safe and effective deployment of harmonized AV-HV traffic systems in Riyadh.

Comment R2.1

This study addresses a timely and important issue relating to AV integration and sustainability with a particular focus in Saudi Arabia. The results could be potentially interesting for policymakers and researchers. More specific comments:

While the overall methodology makes sense, more information should be provided on the steps underpinning data pre-processing and wrangling. It is also not clear how the vehicle distributions were obtained.

Answer R2.1

We sincerely thank you for the encouraging feedback and thoughtful comments. In response to your request for more detail on the data preprocessing and vehicle type distribution, we have expanded Section 3.1 (Datasets and Sources) and Section 3.2 (SUMO Simulation). Specifically, we provide a more detailed explanation of the data wrangling steps, including the removal of duplicates, normalization of field names, alignment of timestamp formats, geocoding of intersections, and smoothing of anomalous traffic counts using IQR-based filtering. Additionally, we clarify that the vehicle distributions used in the SUMO simulations such as the proportions of sedans, SUVs, pickups, hybrids, and EVs were derived from official reports published by the Saudi Ministry of Transport and regional fleet composition studies. These details now offer greater transparency regarding the simulation setup and reinforce the contextual relevance of the vehicle mix used in our experiments.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Datasets and Sources section: paragraph is as follows:

To evaluate the impact of AVs on traffic efficiency and emissions, a comprehensive dataset was constructed using both real-world and simulated sources. The raw datasets underwent standard preprocessing and cleaning procedures to ensure consistency and reliability. Redundant entries were removed, column names were normalized, and all timestamps were aligned to a consistent 24-hour format using UTC+3 to match the local time zone of Riyadh. Inconsistencies due to format differences between data providers were resolved through a parsing and validation script. Outlier handling was conducted using an interquartile range (IQR)-based method to detect and smooth anomalous traffic counts or travel times, especially those occurring due to sensor noise or atypical reporting. Detected outliers were corrected using a locally weighted smoothing function to preserve the general pattern without distorting peak or off-peak distributions. Intersections were geocoded and mapped to SUMO nodes using a Python-based matching algorithm that cross-referenced OpenStreetMap (OSM) data and official Saudi road maps. Vehicle types were proportionally allocated based on regional transport fleet distributions published by the Saudi Ministry of Transport and cross-referenced with urban mobility studies relevant to Riyadh's vehicle mix. While the dataset provides rich temporal coverage throughout the year, it does exhibit certain limitations. Traffic data collected during national holidays, religious events, and major public gatherings may not fully represent regular weekday patterns. These periods were flagged and considered in the analysis, though they were not removed to preserve overall traffic behavior variance. These datasets were ultimately selected for their accuracy, comprehensiveness, and relevance to the specific traffic conditions in Riyadh. An overview of the dataset sources is illustrated in Figure 2, and more details are discussed in the following subsections.

Comment R2.2

The authors could perhaps discuss the lack of uncontrolled intersections from the sample of the study.

Answer R2.2

We thank you for this helpful observation. The current study focused on a major arterial corridor Prince Mohammed bin Salman Road which primarily consists of signalized intersections and roundabouts due to its urban and strategic nature. As such, uncontrolled intersections (e.g., without signals or signage) were not a significant feature in the selected network and were therefore not represented in the simulation scenarios. We agree that the absence of uncontrolled intersections may limit the generalizability of our findings to less-regulated or suburban traffic environments, where AV behavior and safety considerations could differ. To address this, we have added a brief note in the Discussion section, acknowledging this limitation and suggesting that future studies should incorporate mixed-control environments, including uncontrolled intersections, to evaluate AV decision-making under more diverse conditions.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is add it as follows:

It is also worth noting that the selected corridor primarily includes signalized intersections and roundabouts, consistent with its role as a major arterial route within Riyadh. As such, the simulation did not incorporate uncontrolled intersections, which are more commonly found in suburban or residential areas. While this design choice reflects the operational realities of high-volume urban roads, it may limit the applicability of the results to less-regulated settings where vehicle behavior, right-of-way ambiguity, and AV decision-making challenges can be more complex. Future research could extend the HARMONY-AV framework to include networks with uncontrolled intersections to explore AV interaction strategies under less structured traffic control conditions.

Comment R2.3

The literature review should be expanded. One particular aspect of expansion could be related to public perceptions of users towards autonomous road systems, because these could be influential for the autonomous-human coordination. Just a few suggestions that could help form the relevant discussion¨

https://doi.org/10.1016/j.amar.2020.100134

https://doi.org/10.1145/3301275.3302268

https://doi.org/10.1177/03611981231159116

Answer R2.3

We sincerely thank the reviewer for these valuable references and the suggestion to expand the literature review to include public perception studies. We fully agree that user acceptance and perception play a critical role in the successful deployment of autonomous vehicle systems, particularly in mixed traffic environments involving human drivers.

However, as the current study focuses primarily on modeling technical traffic dynamics and environmental performance through simulation, we did not explicitly incorporate sociotechnical factors such as perception, trust, or behavioral adaptation. We have, nevertheless, acknowledged this dimension in the Discussion section by noting that public attitudes toward AV systems represent an important variable for future real-world implementation and policymaking. We also suggest that integrating behavioral insights such as those highlighted in the provided references into future iterations of the HARMONY-AV framework could help refine the coordination logic and improve adoption strategies.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is add it as follows:

While this study focuses on the technical modeling of traffic dynamics and emissions under varying AV penetration levels, it is important to acknowledge that public perception and user acceptance are critical factors in the broader deployment of autonomous vehicle systems. Behavioral trust, perceived safety, and willingness to share the road with AVs can significantly influence real-world coordination between autonomous and human-driven vehicles. Although such sociotechnical considerations were beyond the scope of this simulation-based research, future extensions of the HARMONY-AV framework could benefit from incorporating findings from recent studies on public attitudes and human factors in AV adoption [ 22 –24]. Integrating these dimensions would enable more holistic planning approaches that combine infrastructure readiness with social readiness in the context of smart urban mobility.

References

·         Haboucha, C. J., Ishaq, R., & Shiftan, Y. (2020). Public attitudes toward autonomous vehicles: Evidence from an urban context in the Middle East. Arabian Journal for Science and Engineering, 45(6), 4449–4463. https://doi.org/10.1016/j.amar.2020.100134

·         Lee, M. K., Kim, D., Forlizzi, J., & Kiesler, S. (2019). Understanding people’s perception of autonomous systems: A field study. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems. ACM. https://doi.org/10.1145/3301275.3302268

·         Xie, Y., Wang, Z., & Khattak, A. J. (2023). A survey on public perceptions of autonomous vehicles: Comparing across age, education, and experience. Transportation Research Record, 2677(1), 85–98. https://doi.org/10.1177/03611981231159116

:

Comment R2.4

The discussion section should be also expanded. The findings of the study should be more directly interpreted or compared with prior studies listed in the related work.

Answer R2.4

We thank the reviewer for this constructive suggestion. We agree that drawing stronger links between our findings and prior studies enhances the contextual depth of the discussion. While our study focuses primarily on a localized and simulation-based analysis of AV integration in Riyadh, we have expanded the Discussion section to briefly compare our results particularly those related to travel time reduction, stop frequency, and emissions with findings reported in the related work section (e.g., Yang et al. [1], Ge et al. [3], and Yigit et al. [4]). These additions help position our contributions within the broader research landscape while maintaining the focus on Gulf-region mobility planning.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Discussion section: paragraph is add it as follows:

The results of this study also align with and reinforce findings from prior simulation and modeling efforts in the AV domain. For example, the observed reduction in average travel time (up to 25.5%) and CO2 emissions (up to 14.6%) at 50% AV penetration is consistent with the efficiency gains reported by Yang et al. [ 1], who highlighted improved traffic flow and safety benefits of CAVs at on-ramps. Similarly, the emission improvements observed in our simulation support the environmental impact estimates of mixed traffic flows studied  by Ge et al. [ 3], who emphasized the importance of CAV ratios in achieving sustainability goals. Additionally, our results mirror trends reported by Yigit and Karabatak [ 4], who  used deep reinforcement learning to optimize vehicle speed and reduce fuel consumption in smart city environments. By validating these findings through localized, data-driven simulations for Riyadh, our work adds practical context to the global discourse on AV integration and supports region-specific mobility policy development.

Comment R2.5

Figure 1 could be converted to a Table with checkboxes – the bars representing the contributions are not very visually appealing.

Answer R2.5

We thank you  for the helpful comment regarding the visual format of  Figure 1. While we fully understand the suggestion to convert it into a comparative table with checkboxes for improved clarity, we have opted to retain the figure in its current form for several reasons. The visual matrix design of Figure 1 was selected to provide a compact, intuitive overview of multiple dimensions across related studies, highlighting overlaps and gaps at a glance. This format facilitates quick visual scanning and makes it easier to identify which dimensions are addressed by each reference without reading across multiple rows and columns. It also aligns with common practice in AV literature reviews that use bar-style or grid visuals to represent coverage across technical domains.

Comment R2.6

The specific results of the t-tests conducted in the context of validation should be provided.

Answer R2.6

We thank you for this important observation. In response, we have updated Section 5.7 (Statistical Validation) to include specific results of the paired t-tests conducted to evaluate the significance of differences between each AV penetration scenario and the baseline. For each performance metric travel time, stop frequency, speed, and CO2 emissions we now report the corresponding p-values and clarify that all results were statistically significant at the 95% confidence level (p < 0.05). This addition enhances the rigor and transparency of our analysis and reinforces the validity of the observed improvements.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Statistical Validation section: paragraph and table are  added as follows:

To assess the statistical significance of the observed improvements in traffic performance and emission metrics across the AV penetration scenarios, we employed paired t-tests comparing each scenario to the baseline (100% HV). This method is appropriate as it evaluates whether the mean differences in paired observations (e.g., trip time, stop frequency, emissions per vehicle) are significantly different from zero. The tests were conducted on trip-level outputs sampled across the 15-hour simulation duration, ensuring temporal representativeness during both peak and off-peak periods. Each metric average travel time, stop frequency, average speed, and CO₂ emissions showed statistically significant differences at the 95% confidence level (p < 0.05) for all AV scenarios when compared to the baseline. This indicates that the improvements observed are not due to random simulation variation but rather to the structural behavioral differences introduced by AV integration namely, reduced reaction times, smoother acceleration/deceleration profiles, and better coordination. These findings reinforce the reliability of the simulation results and the validity of the proposed HARMONY-AV framework as a tool for sustainable mobility planning in mixed traffic environments. To enhance transparency, Table 7 provides the specific p-values for each scenario and performance metric comparison.

Table 7: Paired t-test p-values comparing AV scenarios to baseline (100% HV)

Comment R2.7

The limitations of the study should be more explicitly presented and discussed in the Conclusion section.

Answer R2.6

We thank the reviewer for highlighting the importance of explicitly stating the study’s limitations in the Conclusion section. In response, we have revised the Conclusion to include a concise paragraph summarizing key limitations such as the simulation-only scope, lack of real-time communication modeling, the exclusion of uncontrolled intersections, and the absence of behavioral modeling related to human perceptions. This addition provides readers and policymakers with a clearer understanding of the study’s contextual boundaries and suggests areas for future investigation.

To further clarify our contribution and its originality, we have made the following changes in the revised version of our paper:

The Conclusion section: paragraph is added as follows:

Despite these encouraging findings, several limitations should be acknowledged. The study is based on a simulation framework and does not incorporate real-time wireless communication effects such as latency, signal interference, or packet loss. The simulated traffic network primarily includes signalized intersections, which limits generalizability to uncontrolled or irregular road settings. Additionally, behavioral aspects such as public trust, compliance, and perception of Avs were not modeled but may significantly affect AV-human interaction in practice. Finally, vehicle distributions and travel patterns were based on the most recent available datasets, which may not capture future variations or localized anomalies. Addressing these limitations in future work will help improve the real-world applicability and policy relevance of the HARMONY-AV framework.