A humanoid that yields to humans
Agility Robotics has unveiled Digit 5, a new generation humanoid robot engineered from the ground up to work in the same space as humans without safety cages or fences.
When Digit 5 detects a person nearby, its AI-based safe motion system can choose from several mitigations:
- reroute to avoid the person or stop and let them pass;
- if someone comes very close, squat into a seated-like pose to minimize risk;
- adjust speed and trajectory to maintain a safe buffer.
Agility calls this approach “cooperative safety”: robots and humans share the same workspace while keeping a safe distance.
Sensors, AI compute and safety standards
Digit 5 uses an onboard array of multimodal, vision-based sensors plus AI algorithms to detect and track humans in its vicinity. For compute, Agility relies on Nvidia Thor IGX hardware and the Nvidia Halos for Robotics safety software stack. The company is among the first to push Thor IGX into a functionally safe industrial application.
In parallel, Agility has been involved in drafting ISO 25785-1, an upcoming international safety standard for industrial mobile robots. The goal is a more granular safety response around humans, instead of simply powering robots off whenever people come close.
Built for real warehouse and factory work
Digit 5’s redesigned legs can repeatedly lift up to 50 lb (≈23 kg), covering all single-person lift tasks under US OSHA guidelines. Key specs:
- Height 5'11" (about 180 cm), with 7.2 ft (≈2.2 m) reach – similar to an average adult’s reach to warehouse shelves
- 90 minutes runtime per charge, with 9-minute fast charging
- Over 20 hours of productive work in a 24-hour period
- Quick-swap grippers / end-effectors for different manipulation tasks
This lets Digit 5 go beyond simple tote handling to tasks like palletizing, packaging and moving through busy aisles where people also work. Once cooperative safety is proven at scale, Agility sees this as the foundation for “collaborative safety”, where humanoids and humans work side by side in close proximity.
Source: Ars Technica
