Bridging the Gap between Deep Learning and Probabilistic Inference for Advancements in Robotics

Dear Colleagues,

In recent times, the field of robotics has witnessed remarkable advancements propelled by the integration of artificial intelligence (AI) and machine learning (ML), particularly deep learning, into robotic systems. These developments have propelled robots to accomplish tasks with unprecedented levels of performance. This has spurred a reconsideration of the traditional probabilistic inference algorithms that have long been relied upon for reliable operation in uncertain and unstructured environments. These probabilistic techniques offer a comprehensive framework that unifies perception, control, and learning in robotics.

Concurrently, the paradigm of robot learning has gained significant momentum. It holds the promise of enabling robots to generalize their capabilities across a spectrum of scenarios, mitigating the necessity for the meticulous engineering of task-specific models, which is a hallmark of classic probabilistic methods. Yet, a fundamental question persists: can we entrust robots with dependable and adaptive behaviors solely through data-driven learning approaches?

Addressing this question lies at the heart of this Special Issue: Bridging the Gap between Deep Learning and Probabilistic Inference in Robotics. As the robotics landscape evolves, there is a growing recognition that extracting the best elements from both deep learning and probabilistic inference could hold the key to unlocking new levels of robotic performance.

This Special Issue serves as a platform to delve into this juncture where robotics, deep learning, and probabilistic inference converge. It aims to foster dialogue among researchers who are navigating the complexities of merging these diverse methodologies. By uniting experts from these distinct fields, this issue aims to push the boundaries of what is possible in intelligent robotics.

Key Objectives and Themes:

• Learned Components in Probabilistic Contexts: A recent breakthrough in this domain has been the introduction of learned components within probabilistic frameworks. This issue seeks to explore the potential of such integration to enhance system adaptability.

• Unpacking Epistemic Uncertainty: Epistemic uncertainty, often referred to as model uncertainty, encapsulates what we do not know about the underlying data distribution. It arises when the available data are insufficient to fully characterize the intricacies of the environment. In the context of robot learning, this uncertainty could manifest as gaps in the training data, variations in sensor measurements, or unmodeled dynamics of the robot and its surroundings.

• End-to-End Differentiable Algorithms: The emergence of end-to-end differentiable algorithms is a significant milestone. This issue will delve into the implications of these algorithms on robust and reliable learning systems.

• Bayesian inference for Probabilistic Reasoning: Recent advancements have propelled Bayesian inference to new heights within the field of robotics. This paves the way for seamlessly integrating learning and reasoning, potentially bridging the gap between the adaptability of deep learning and the robustness of probabilistic reasoning.

• Adaptability for Robust Control: Adaptability in an integrated robotic system is often the most crucial factor to ensure robustness, even in the mist of uncertainty. The ultimate goal is to cultivate robotic learning systems that not only learn from data but also adapt to novel situations effectively.

• Learning to Plan and Planning to Learn: The integration of robot learning and planning is a two-way street. On the one hand, robots can utilize data-driven insights to make more informed decisions and optimize action sequences. On the other hand, planning can be employed to facilitate the learning process. Robots can devise deliberate actions to gather informative data, thereby accelerating the learning curve and refining their understanding of the environment.

Dr. Tin Lai
Prof. Dr. Zhaojie Ju
Guest Editors

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Electronics is an international peer-reviewed open access semimonthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 2400 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• robot learning
• deep learning in robotics
• probabilistic robotics