Back to articles
Robotics & Physical AI

Agility’s Digit 5 Can Stop or Squat to Keep Human Coworkers Safe

3 min read

Introduction

For humanoid robots entering warehouses and factories, the difficult question is not only whether they can move boxes. It is whether they can share a changing workspace with people without requiring every machine to sit inside a cage. Agility Robotics is positioning its new Digit 5 around that problem. When the robot detects a person nearby, it can move out of the way or stop to let the person pass. If someone comes especially close, Digit 5 can lower itself into a squat or seated posture to reduce the potential for a harmful contact.

Key points

  • Graduated responses instead of a simple shutdown. Agility says Digit 5 uses a safe-motion system that selects different mitigations based on the location and proximity of a person. The goal is to make safety more nuanced than switching the robot entirely on or off.
  • Perception is central to the safety case. The robot uses AI algorithms and several types of vision-based sensors to monitor nearby people. Its onboard computing is built around Nvidia’s Thor IGX hardware and incorporates the Nvidia Halos software system for robotics safety applications. Agility has not disclosed the specific sensor models.
  • A broader industrial task set. An upgraded leg design allows Digit 5 to repeatedly lift loads of up to 50 pounds. Its battery provides 90 minutes of operation and can recharge in about nine minutes. The robot is also taller, reaches higher shelves, and supports swappable grippers for tasks such as palletizing, packaging, and material handling.
  • A staged launch. Agility expects to give selected customers early access during the first half of 2027 and make the robot generally available by the end of that year. The company is retooling its RoboFab facility in Salem, Oregon, for production.

Why it matters

Digit 5’s most important feature may be its operating model rather than its payload. Industrial robots have traditionally been placed in guarded cells, with people kept out of the work area or the equipment stopped when someone enters. That arrangement is difficult to apply to warehouse aisles, pallet exchange, and mixed human-machine workflows. A robot that can pause, yield space, or lower its body could make less segregated deployments more practical.

Agility says earlier Digit models have accumulated more than 65,000 hours in warehouses and factories across North America. Customers and pilot users have included GXO, Schaeffler, Amazon, and Toyota Motor Manufacturing Canada. That experience provides operational context, but it does not by itself establish that Digit 5 will be safe in every open, crowded, or visually difficult environment. Its behavior will still need to be assessed across real lighting conditions, traffic patterns, and workplace procedures.

Agility describes the initial approach as “cooperative safety”: humans and robots share a workspace while the robot maintains a safe distance. If that model proves reliable in commercial deployments, the company sees it as a foundation for “collaborative safety,” in which people and humanoids work directly alongside one another at close range. Agility is also participating in work on ISO 25785-1, an international safety standard for industrial mobile robots that remains under committee review. Digit 5 is therefore both a product upgrade and a test of how much safety infrastructure humanoid robots need before open workplaces can become the norm.

Source: Ars Technica AI

Comments

Checking sign-in status...

Loading comments...

Related articles

CCTest · Blog
PhysBrain 1.5 Unifies Perception, Action, and Future-State Prediction
Robotics & Physical AI
cctest.ai

PhysBrain 1.5 Unifies Perception, Action, and Future-State Prediction

PhysBrain 1.5 extends a vision-language model into a unified embodied foundation model that can interpret physical scenes, generate end-effector motion, and predict subsequent states. The paper reports an average score of 72.5 across 28 embodied understanding benchmarks for its 8B model.

Read more
CCTest · Blog
MobileVLA-R1 2.0 Links Robot Reasoning to Action with RL
Robotics & Physical AI
cctest.ai

MobileVLA-R1 2.0 Links Robot Reasoning to Action with RL

MobileVLA-R1 2.0 combines structured embodied reasoning, reinforcement learning, and a reasoning-conditioned action decoder to narrow the gap between language understanding and reliable robot execution. The framework is evaluated across navigation, quadruped control, and humanoid mobile manipulation.

Read more