AGIBOT G2 Logs 1,730 Store Tasks

This is a valuable field benchmark, but not yet proof of autonomous retail: the 4-of-6 pass rate and missing teleoperation data expose how far embodied AI remains from dependable deployment.

• –A live customer-facing store tests adaptability more rigorously than scripted laboratory demos.
• –Task counts are encouraging, but greetings and complex food preparation are not equivalent measures of performance.
• –Genie Master’s planning, execution, feedback, and training-data loop may matter more than the robot hardware itself.
• –The failed assessments, reportedly involving cleaning and shelf operations, highlight the difficulty of handling edge cases and changing environments.
• –Developers need intervention logs, uptime, charging data, and per-task success rates before comparing this trial with commercial automation benchmarks.