I gave my robot physical memory – it stopped repeating mistakes

A new robotics memory system called robotmem enables robots to store and retrieve past experiences, preventing them from starting fresh with each task. The system records parameters, trajectories, and outcomes from previous attempts, then uses hybrid search to find relevant memories for future decisions. Testing on a FetchPush experiment showed a 25% improvement in success rates, jumping from 42% to 67% completion. Unlike existing memory systems designed for text or academic models, robotmem specifically handles physical robot data including spatial information and mechanical parameters.