Galbot G1 Review (3★)

Editorial

The Galbot G1 earns its place as the first robot to secure a licensed pharmacy operating permit in China and run continuous shifts at FamilyMart, proving wheeled dual-arm hardware plus end-to-end VLA models can move beyond demos into regulated retail. Yet its 5 kg/arm payload and 1.5 m/s speed deliver only marginal productivity gains when the system masters fewer than 1% of SKUs and still needs four humans per shift. The 8-hour runtime and 40% grasp failures on common items expose brittle sim-to-real transfer even with GraspVLA pretraining on a billion synthetic frames. At ~¥700,000 per unit this is a credible early commercial platform for narrow Chinese retail and pharmacy workflows, but it remains far from general-purpose AMR replacement. Deployment scale is also modest—under 150 units/year output against hundreds of orders—highlighting that regulatory wins outpace manufacturing maturity. For operators seeking reliable shelf stocking today, the G1 is a specialized tool, not a benchmark.

Pros

• Continuous daily operation in a real FamilyMart store in Beijing's Zhongguancun district, first such ongoing retail deployment for an intelligent robot
• 5 kg payload per arm (10 kg combined) with dual-arm configuration including suction and adaptive gripper end-effectors
• 8-hour laboratory runtime on 48V 30Ah battery with NVIDIA Jetson AGX Orin compute supporting AstraBrain/GraspVLA models
• Maximum speed of ~1.5 m/s on omnidirectional wheeled base with 1.9 m horizontal reach and torso lift to 2.4 m

Cons

• Handles fewer than 10 of ~1000 store SKUs in live FamilyMart testing with ~40% grasp failure rate on chip bags
• Requires four human staff still present in deployed stores; pharmacy units have not learned full medicine locations
• 8-hour lab runtime falls short of 24/7 claims without frequent charging; real deployments remain curated task sets
• 21 articulated joints but limited to wheeled base, no bipedal mobility, and narrow autonomy outside retail picking