Humanoid robots can only scale if they can operate safely alongside people. Pras Velagapudi, CTO of Agility explains how NVIDIA Halos is helping make that possible.

Learn more about physical AI safety:

Safety at scale requires both local perception and absolute spatial ground truth. Local safety stacks like Halos solve real-time obstacle avoidance, but centimeter-level RTK positioning is the other critical piece of the puzzle—giving Physical AI systems ground-truth spatial awareness in the real world. 🤖

@agilityrobotics $CCXI

@agilityrobotics need this those bot fight videos are having me nervous fr

@agilityrobotics Pump it boys. Let's go. $CCXI

@agilityrobotics the safety side of humanoid robots is gonna matter so much once they’re actually around people every day.. making them predictable in real-world spaces is a whole different challenge😳

@agilityrobotics Move fast and break things gets less charming when the software has elbows.

Safety onboard is necessary. It is not the plant floor problem. Industrial robots have run predictable paths, known cycle times, and known faults for 50 years. Humanoid robots will create new problems, unstructured motion, lower robustness, and output that still cannot beat a custom cell built with a real factor of safety. The skill gap is already bad. Who clears a joint fault, a dropped part, or a model doing something a PLC controlled machine never would at 2am? Until they diagnose and troubleshoot themselves, they will sit faulted waiting on a specialist most plants do not have. That is not scale. That is a new manufacturing tax.