About

Welcome to the CoRL 2023 Workshop on Bridging the Gap between Cognitive Science and Robot Learning in the Real World: Progresses and New Directions!

Humans possess remarkable cognitive abilities that enable them to solve everyday tasks and adapt to new situations. These cognitive functions operate at different levels, including the subconscious (e.g., perception and action), conscious (e.g., planning and reasoning), and introspective (e.g., self-monitoring and self-explanation). Importantly, these levels of cognition are not separate entities but are intricately intertwined and mutually adaptive. Advanced agents like humans rely on higher-level cognitive functions, such as introspection and self-explanation, to guide and inform their lower-level cognitive processes, such as planning and control.

In contrast, current robots lack the conscious and introspective human-level cognitive intelligence that can effectively guide their subconscious intelligence. This absence limits their ability to efficiently perform complex tasks, adapt to novel scenarios, and seamlessly interact with humans. Granting robots similar cognitive intelligence holds significant benefits, including the potential to enhance their capacity for learning, facilitate their adaptation to unforeseen circumstances, and enable more natural and intuitive interactions with human counterparts. One promising hypothesis to achieve this is to establish strong connections between principles derived from cognitive science literature and robot learning.

This workshop seeks to stimulate productive discussions on the underexplored connections between human cognitive procedures and robot learning techniques. By exploring these intersections, we aim to explore new research topics and define new exciting research agendas in the field of robot learning. The workshop will emphasize the identification of new research directions that draw inspiration from cognitive science. The ultimate goal is to enable robots to learn and operate more effectively in real-world settings, potentially alongside humans.

Topics of Interest:
 •  Robot learning guided by self-explanation or self-introspection: Investigations into methods and algorithms that enable robots to acquire self-explanatory and introspective cognitive abilities, allowing them to understand and improve their own learning processes
 •  Leveraging world models for robot learning: Approaches that utilize world models, including simulated environments or predictive models, to facilitate robot learning and enhance their adaptability to new scenarios and tasks
 •  Robot learning with language feedback and/or visual attention: Exploration of techniques that incorporate language feedback or visual attention mechanisms to improve robot learning and enable more effective interaction with the environment and humans
 •  Learning through interaction with humans: Studies on the integration of human-robot interaction and collaborative learning paradigms, aiming to enhance the robot's learning capabilities through direct engagement and cooperation with human partners
 •  Robot learning for long-horizon tasks: Investigations into long-term learning approaches that enable robots to acquire and refine skills over extended periods, enabling them to handle complex, multi-stage tasks and maintain performance over time
 •  Augmenting robot learning using long-term (e.g., physical) memories: Research focused on leveraging long-term memory mechanisms, such as physical storage or other memory structures, that enable enhanced robot learning. This includes investigating approaches that facilitate efficient retrieval of stored knowledge, as well as the ability to update and integrate new information into the existing memory, enabling robots to improve their performance, adaptability, and decision-making in complex scenarios
 •  Learning multi-robot collaborations: Advancements in the area of multi-robot systems, including cooperative learning, task allocation, communication, and coordination, to enable robots to collaborate and learn from each other in a team-based setting
 •  Utilizing large language models for improving robot learning: Exploring the application of large language models, such as GPT, to enhance robot learning by leveraging natural language processing and understanding capabilities