AGIBOT Reports Collective Learning Experiments with 96 Robots

In a November 2 announcement, AGIBOT said it had released Collective Experience Learning (CEL) with research partners on October 30. The approach uses successful and failed robot trajectories as shared learning material.

AGIBOT reports 600 real-robot tests covering flower arranging, clothes folding, phone charging, stool assembly and phone packaging, with nearly 30 robot-hours of testing. It reports an overall task success rate of 97.5% for CEL FERT.

The company says its fault-tolerant asynchronous training system let individual robots recover without resetting the whole cluster. In its 96-robot experiment, it reports 9,834 task-collection rounds and 30 robot faults, with faults accounting for 1.3% of total cluster operating time.

← More reports