Proximal State Nudging reduces AI-assisted skill atrophy by 7x

A team of researchers from Stanford University, Toyota Research Institute, and other institutions has published a new paper on a shared autonomy algorithm called Proximal State Nudging (PSN). The algorithm addresses skill atrophy—the gradual decline of human capability when relying too heavily on AI assistance—which poses safety risks in semi-autonomous systems like self-driving cars or robotic teleoperation. PSN works by nudging users toward states that are estimated to be most learnable, balancing task performance with long-term human skill development. The team first tested PSN using simulated students in the classic LunarLander environment, showing it outperformed existing shared autonomy baselines in balancing student improvement and system performance.

In what the authors call the first human subject studies of a planner incorporating learning-compatible shared autonomy, the team tested PSN across two driving tasks in the CARLA simulator (high-performance racing and parallel parking, n=60). Results showed PSN produced up to 7x larger gains in unassisted skill compared to standard blended shared autonomy, while incurring 50% fewer collisions than unassisted self-practice. This suggests PSN can help operators build genuine competence even while using AI aids, potentially reducing overreliance and improving safety in high-stakes domains like autonomous driving, drone piloting, and surgical robotics.