Memory as Plans: World-Action Modeling with Memory-Grounded Planning
MaP-WAM turns a robot’s task history into compact plans, keeping execution latency roughly constant as the history grows.
The paper targets manipulation tasks where the current observation is not enough and long-horizon memory matters. Its framework stores completed task segments as language instructions plus sparse visual context, then uses that episodic memory to produce the next language plan and visual guidance. A World-Action-Progress model executes each plan for an unknown duration, predicting both action chunks and progress so it can shift segments adaptively. The authors report 83.3% success on RMBench and 78.0% success on real-robot tasks. HF Daily Papers' note
The paper targets manipulation tasks where the current observation is not enough and long-horizon memory matters. Its framework stores completed task segments as language instructions plus sparse visual context, then uses that episodic memory to produce the next language plan and visual guidance. A World-Action-Progress model executes each plan for an unknown duration, predicting both action chunks and progress so it can shift segments adaptively. The authors report 83.3% success on RMBench and 78.0% success on real-robot tasks. HF Daily Papers' note
score 5