2nd Workshop on Trends and Advances in Machine Learning and Automated Reasoning for Intelligent Robots and Systems

2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022)
October 27th, 2022, Kyoto, Japan

Scope


In recent years, data-driven machine learning (ML) methods have achieved good performance in various application problems. For example, deep neural networks have demonstrated their ability to deliver state of the art performance in domains such as computer vision, natural language processing, robot perception and control, and human-robot collaboration. Despite these advances, machine learning models are often block boxes that require many labeled training examples to construct, may lead to erroneous outcomes, and whose internal operation is difficult to interpret. This is due to multiple factors such as biases in data, errors in the domain model or data samples, incorrect interpretation of algorithm outputs, and lack of context in decision making. Knowledge representation and reasoning (KRR) methods, on the other hand, have a well-established history of supporting different types of reasoning, including reasoning with incomplete commonsense domain knowledge. It is, however, not feasible to provide a comprehensive encoding of domain knowledge in complex domains, and the encoded knowledge may need to be revised over time.

The complementary strengths of ML and KRR methods can be exploited to address the challenges highlighted above. In particular, KRR methods and knowledge provided by domain experts can be used to constrain the search space and direct the learning and revision of predictive models using ML models. The concepts and rules acquired by the ML methods can, in turn, be used to augment or revise the knowledge used by the KRR methods. There has been considerable research on developing KRR tools, e.g., Description Logics, Answer Set Programming, Semantic Web Languages which can be integrated within hybrid frameworks for enhancing robot's perception, control and tasks planning capabilities. Despite considerable research in recent times, e.g., on neuro-symbolic methods, an effective combination of ML and KRR methods remains an open problem. Creating methods and frameworks that address this open problem can help us fully exploit the capabilities of AI systems in general. It can also lead to robot systems that can be proactive and resilient to changes, automatically adapt reasoning and learning to different situations, and build sophisticated anticipatory models of intentions, behaviors, contingencies to support effective collaboration with humans in complex domains.

The proposed workshop intends to bring together researchers from different sub-fields of AI to discuss how best to address challenges in ML and KRR for intelligent robots and systems. Invited speakers will present recent advancements in ML methods, KRR methods, hybrid frameworks that combine ML and KRR, and the application of these methods and frameworks to practical problems in different domains. We also seek to explore important related topics such as standardization and ethics in intelligent robots and systems.

Endorsement


This workshop is endorsed by the following IEEE RAS Technical Committees:

Supported by