Research · Robotics & embodied AI
Synthetic demonstrations let robot policies escape sparse-reward failures
SynthDemo-RL: Breaking the Zero-Reward Barrier in VLA Adaptation with LLM-Guided Synthetic Demonstrations
An automated teacher supplied successful manipulation trajectories that enabled subsequent reinforcement learning on tasks where direct PPO received no reward.
LIBERO-PRO: 97.8% success rate (as reported)
Fujitsu Limited, The Institute of Statistical Mathematics