Heng Yang

Forcing all memory into policy weights fundamentally bottlenecks robot manipulation: offloading persistent object identity and containment to an external, LLM-queried ledger jumps benchmark success from 45.9% to 64.3%.

This work proposes RoboMP-DINOv2 (Robotics Mask-Prompted DINOv2), a full-scene vision encoder that treats masks as spatial prompts rather than visibility filters, and introduces masked-region color randomization (MCR) to improve appearance robustness, yielding RoboMP-DINOv2-MCR.