post-training vision-language-action policies for industrial disassemblyExpert
PACER: Process-Aware VLA Policy Learning from Human Corrections for Robotic Disassembly
A Texas A&M-led team shows that a robot policy pulling components off computer motherboards gets far better when its own failed attempts and an operator's rescues are weighted by how close each move got to the target. Success on 100 unseen hardware trials rose to 87 from 54 with ordinary fine-tuning, which matters for anyone trying to automate e-waste recycling with little training data.