[Robot Learning Team Seminar] Guest Talks by Prof. Giovanni Beltrame and Prof. Glen Berseth (Mila)

The RIKEN AIP Robot Learning Team hosts two guest talks from Mila researchers on robot learning and physical AI. Online participation via Zoom is open to all registered attendees, while on-site attendance at the RIKEN Nihonbashi Office is limited to RIKEN members. The two talks run back to back in the late afternoon. Prof. Giovanni Beltrame (Polytechnique Montréal, Mila) presents "Physical AI for Field Robotics: Challenges and Applications." He examines how learned models can be made reliable in unstructured environments such as search-and-rescue sites, agricultural fields, and underground mines, covering perception and mapping without GPS, coordination without infrastructure, resource constraints, and the failure modes of large pretrained vision models under distribution shift. Prof. Glen Berseth (Université de Montréal, Mila) presents "From Trial and Error to Foresight: Building Robots That Learn, Generalize, and Act." He discusses how robots learn from experience, build structured visual representations, and generalize across tasks, along with making large pretrained robot control models more capable through planning and data relabeling, and using large language models to generate reward signals automatically.