Physical Intelligence Unveils MEM Architecture for Long-Horizon Robotic Memory
ycombinator · x · 2026-08-08
Physical Intelligence (π) presented MEM (Multi-Scale Embodied Memory) at the Y Combinator Paper Club, targeting the memory bottleneck of Vision-Language-Action (VLA) models in long-horizon tasks.
- Core Problem: While current robotic foundation models master individual skills like washing pans, they struggle with multi-stage tasks like cleaning a kitchen due to a lack of coherent memory.
- Memory Challenge: Retaining full observation history causes context explosion, but discarding it leads to "causal confusion," exacerbating spurious correlations in imitation learning.
- Solution: The MEM architecture introduces multi-scale long and short-term memory for VLA models, enabling robots to track task progress, remember out-of-view objects, and recall past trial-and-error outcomes.
More from Embodied
- openpilot brings stoplight-stopping autonomy to a 2019 Corolla — k7agar · 2026-10-16
- Hironobu Fujiyoshi publishes deck: from AI agents to Agentic Robotics — HirokatuKataoka · 2026-10-16
- CoDA co-evolves simulation and robot policies via shared failure memory, lifting OOD success from 5.3% to 27.6% — PLAN-Lab · 2026-10-16
- Beff Jezos claims rocket Cybertruck will be the first Tesla/SpaceX lunar vehicle — beffjezos · 2026-10-16
- Maglev 6D shuttles levitate under 6mm, hit 5-10µm precision, avoid collisions in firmware — lukas_m_ziegler · 2026-10-16
- Robotic-arm self-shopping cart concept imagines grabbing groceries off the shelf — djcows · 2026-10-16