Back
Physical AI · VLA · RL

VLA for Sidewalk Robot (COCO)

COCO sidewalk delivery robot

Research on enabling more intelligent driving for COCO, a sidewalk delivery robot, by building a Vision-Language-Action (VLA) policy that grounds camera observations and language instructions directly into navigation actions. The policy is refined with reinforcement learning (RL) and real-world driving data so the robot can handle crowded sidewalks, crossings, and unexpected obstacles in urban delivery environments.

Keywords

  • Vision-Language-Action (VLA) models for end-to-end sidewalk navigation
  • Reinforcement Learning (RL) fine-tuning of driving policies
  • Physical AI deployed on a real delivery robot
  • Urban navigation among pedestrians, crossings, and dynamic obstacles