MPCoT: Reward-Guided Multi-Path Latent Reasoning for Test-Time Scalable Vision-Language-Action

Benchmark Model Rank Results
robot-manipulation-on-calvinMPCoT#1avg. sequence length (D to D): 4.79
robot-manipulation-on-libero-spatialMPCoT#7Success Rate: 99.5
robot-manipulation-on-liberoMPCoT#12Success Rate: 99.0
robot-manipulation-on-libero-goalMPCoT#13Success Rate: 99.0
robot-manipulation-on-libero-longMPCoT#16Success Rate: 97.8
robot-manipulation-on-libero-objectMPCoT#22Success Rate: 99.8