Event
Dream2Assist: Intent-Conditioned World Models for Shared Control
1 reports · 1 sources
Jonathan DeCastro and co-authors released a paper proposing Dream2Assist—a model-based reinforcement learning framework that incorporates intent inference into a recurrent state-space world model, enabling assistive policies to anticipate and support humans' discrete goals without taking over human control. The authors report that intent conditioning yields measurable improvements in torque alignment and collision rate. The authors address the simulation-to-real-human gap through policy screening and reward restructuring in iterative human-machine interface design, and present a deployment pipeline from world-model policies to safe human-machine systems. Validation includes two human-subject studies—high-speed racing (N=30) and urban driving (N=10)—as well as a preliminary real-vehicle deployment.
Report timeline
- Dream2Assist: Intent-Conditioned World Models for Shared Vehicle Control
The authors propose Dream2Assist, a model-based reinforcement learning framework that adds intent inference to a recurrent state-space world model, enabling an assistive policy to anticipate and support discrete human goals without taking over human control. The work also presents a complete deployment pipeline from world-model policy to a safe human-machine system, addressing the sim-to-real-human gap through policy down-selection and reward restructuring, and validates it in two human-subjects studies, high-speed racing (N=30) and urban driving (N=10), plus a preliminary on-vehicle deployment, where intent conditioning produced measurable improvements in torque alignment and collision rate.
arXiv cs.RO ·