Memory grounded planning improves real robot manipulation tasks
Memory as Plans: World-Action Modeling with Memory-Grounded Planning
Robotics
Summary
Robots often struggle with tasks that require remembering lots of past details, because their usual methods only look at what’s happening right now. The authors present MaP-WAM, a way for robots to store past experiences as short plans, which helps them remember important details without getting overwhelmed. This new approach lets robots plan better and act more precisely over longer tasks, making memory use more efficient and improving success rates both in tests and on real robots. Their system keeps the amount of information the robot uses during actions fixed, so it stays fast even as tasks get longer.
What this means in practice
- •For robotics engineers: Improve long-horizon task execution in robots by using memory-grounded plans that maintain efficient context length during actions.
- •For automation system designers: Develop more reliable automated manipulation systems that adapt to changing tasks with structured memory and adaptive progress calibration.
Authors
Sizhe Zhao, Haozhe Xie, Weiyu Zhao, Chenchu Zhang, Huan Wang, Chenyang Wang, Qinglin Liu, Shengping Zhang
Abstract
Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their combinations, and may therefore lose fine-grained visual evidence or face a trade-off between history coverage and execution efficiency. We introduce MaP-WAM, a Memory-as-Plans framework that decomposes memory-dependent world-action modeling into memory-grounded planning and plan-conditioned execution, and uses long-term multimodal episodic context as planning-time evidence rather than repeatedly conditioning the executor on the full history. MaP-WAM represents memory as completed segment records containing language instructions and sparse visual context, and converts this episodic memory into compact plans comprising the next segment-level language plan and corresponding visual guidance. A World-Action-Progress (WAP) model executes each plan over an unknown duration by jointly predicting action chunks and corresponding execution progress at inference time, calibrating predicted progress through plan-observation alignment for adaptive segment transitions and closed-loop context updates. MaP-WAM keeps the executor context length fixed, while structured attention further enables key-value caching in both planning and execution. MaP-WAM achieves state-of-the-art performance on RMBench with an 83.3% success rate and attains 78.0% success on real-robot tasks, while maintaining approximately constant executor inference latency as task history grows.