Stanford Robotics Seminar ENGR319 | Spring 2026 | Interactive Autonomy

Why It Matters

Fast, socially aware multi‑robot planning makes autonomous systems viable in everyday settings, reducing accidents and boosting operational efficiency.

Key Takeaways

• Robots need joint prediction and planning for safe multi-agent interaction.
• Potential games reduce multi-robot equilibrium computation to a single optimization.
• Experiments show 20× speedup over traditional game solvers.
• Social conventions affect robot coordination; adapting to norms is crucial.
• Constrained multi-robot tasks demonstrate real-world applicability of game-theoretic methods.

Summary

The Stanford Robotics Seminar focused on interactive autonomy, emphasizing the need for robots to interact safely and intelligently with humans and other agents across domains such as warehouses, manufacturing, and drones. The speaker highlighted that successful interaction requires joint prediction and planning—what economists call a "theory of mind"—so robots can anticipate how others will react to their actions. Key insights centered on framing multi‑robot coordination as a dynamic game. While Nash equilibria capture the ideal joint‑planning solution, computing them in real time is intractable for nonlinear systems. The lab discovered that many practical scenarios fit the class of potential games, allowing a single optimal‑control problem to replace coupled game solvers, yielding up to a twenty‑fold speed increase in experiments with quadcopters and multi‑drone transport tasks. Illustrative examples ranged from a restaurant robot that went rogue, to Whimmo cars stuck in San Francisco traffic, and Amazon warehouse bots deadlocked in a standoff. The team demonstrated that two quadrotors carrying a rigid rod could reorient themselves around humans using fast receding‑horizon game‑theoretic planning. A personal anecdote about differing pedestrian yielding conventions in Singapore underscored the importance of robots recognizing and adapting to local social norms. The implications are clear: by leveraging potential‑game reductions, robots can achieve real‑time, constraint‑aware coordination, opening the door to reliable deployment in complex, human‑shared environments. Moreover, embedding cultural conventions into decision‑making will be essential for widespread acceptance and safety of autonomous systems.

Original Description

For more information about Stanford’s graduate programs, visit: https://online.stanford.edu/graduate-education

May 1, 2026

This seminar covers:

• Learning and control for multi-agent interactions

• Game-theoretic planning and control for robots

• Learning in interactive domains, including imitation learning and reinforcement learning

Follow along with the seminar schedule, visit: https://stanfordasl.github.io/robotics_seminar/

Negar Mehr is an assistant professor in the Mechanical Engineering Department at the University of California at Berkeley. She runs the ICON Lab.