Reinforcing Action Policies by Prophesying

Benchmark Model Rank Results
robot-manipulation-on-simplerenv-widow-xPi0.5-3B + FA-GRPO & FlowScale#42Average: 0.510
robot-manipulation-on-libero-spatialVLA-Adapter + FA-GRPO & FlowScale#86Success Rate: 94.6
robot-manipulation-on-libero-longVLA-Adapter + FA-GRPO & FlowScale#89Success Rate: 86.4
robot-manipulation-on-libero-goalVLA-Adapter + FA-GRPO & FlowScale#93Success Rate: 91.2
robot-manipulation-on-libero-objectVLA-Adapter + FA-GRPO & FlowScale#101Success Rate: 87.4
robot-manipulation-on-liberoVLA-Adapter + FA-GRPO & FlowScale#125Success Rate: 90.1