LIFT Framework Adds Reactive Force Injection to VLA Policies

LIFT implements a reactive action expert alongside pretrained Vision-Language-Action policies to handle contact-heavy manipulation. It uses causal force memory and zero-initialized cross-attention to inject 6D end-effector force, mitigating execution errors caused by occlusion or depth ambiguity.