Self-Improving Vision-Language-Action Models with Data Generation via Residual RL
Open paper
Benchmark
Model
Rank
Results
robot-manipulation-on-libero-goal
OpenVLA w/ PLD
–
Success Rate: 98.9
robot-manipulation-on-libero-object
OpenVLA w/ PLD
–
Success Rate: 99.1
robot-manipulation-on-libero-spatial
OpenVLA w/ PLD
–
Success Rate: 99.5
Rank counts only results with a code link.