Tech • AI • Robotics • Game
This New AI Robot Just Crossed a Huge Line: Digit 5
9.4/10TL;DR
Humanoid robotics is moving from staged demos to industrial deployment, with Agility Robotics, major Chinese manufacturers, and Unitree all showcasing rapid advances in safety, production, and autonomy.
KEY POINTS
Agility unveils Digit 5 for cage-free work
Agility Robotics introduced Digit 5, a 284-pound humanoid designed to work directly alongside people in warehouses and factories without safety fencing. The company calls the approach cooperatively safe work at scale and says the robot was built from customer requirements gathered over three years of operating Digit 4. Agility says it already holds more than $300 million in multi-year orders.
Safety strategy centers on stopping, signaling, and sitting
Digit 5 uses a three-layer safety system: human detection, motion and audio cues, and an independent safety controller. If a person gets too close, the robot can avoid them, stop, or lower its load and drop into a stable seated squat, a notable attempt to address the hazard of a falling biped. Agility is also working through US, Canadian, and international safety standards and has partnered with Nvidia and Fort Robotics to support certification.
Hardware shifts from bird-like legs to lifting performance
Earlier Digit models used backward-jointed legs inspired by animal locomotion, but Digit 5 replaces them with a more conventional design optimized for repetitive lifting. It can repeatedly lift 50 pounds or 22.7 kilograms, about 40% more than Digit 4, and reach 7.2 feet or 2.2 meters high. Its end effectors can also be swapped using standard mounts, allowing it to move from tote handling to machine tending.
Shorter runtime, faster charging, longer daily output
Battery performance takes a different path than many mobile robots. Digit 5 runs for about 90 minutes per charge, down from roughly four hours on Digit 4, but recharges itself in only nine minutes. Agility says that 10:1 run-to-charge ratio allows roughly 21 productive hours a day with around four hours of charging breaks.
Commercial record gives Agility an edge
Agility argues its advantage is not spectacle but operating history. Digit 4 has logged more than 65,000 operational hours across customers including GXO, Amazon, Toyota Motor Manufacturing Canada, and Schaeffler. At GXO’s Flowery Branch site near Atlanta, the robot handled 100,000 cumulative totes by November 2025 at about 98% accuracy.
Public-market push and cost targets
Agility’s merger with Churchill Capital Corp. XI values the company at about $2.5 billion and is expected to raise more than $620 million, largely for scaling production. SEC filings indicate an initial bill of materials of roughly $150,000 to $200,000 per robot, with a target of under $50,000 at 10,000 units a year. The company’s illustrative customer price is about $8,500 a month, pitched as roughly equivalent to 2.5 human shifts.
US factory output meets rising Chinese competition
Agility’s RoboFab plant in Salem, Oregon, was built for up to 10,000 robots a year. In China, UBTech Robotics and Siemens opened a 14,000-square-meter humanoid factory in Liuzhou, Guangxi, also targeting 10,000 units annually, or one robot every 10 minutes. The site uses robots for loading, palletizing, logistics, and inspection, underscoring China’s bid for scale in humanoid manufacturing.
China pairs manufacturing scale with control advances
Chinese firms are also pushing autonomy. AgiBot launched Agile 2.0, a locomotion system that places visual perception directly into the robot’s motion-control loop so it can replan posture, contact, and trajectory in real time. Demonstrations included coordinated rope jumping, balancing on a rolling ball, juggling, and collaborative box handling, all aimed at reducing scenario-specific programming.
Unitree tests fully autonomous combat behavior
On September 7, Unitree showed its G1 humanoid sparring with a human trainer while operating without a visible headset, controller, or teleoperator. The company described the system as a real-time, world-model-driven autonomous combat robot able to predict near-future motion for planning strikes, slips, and footwork. The demo still raised open questions about offboard computing, latency, and how the system performs against a resisting opponent rather than a compliant pad holder.
The industry is shifting from demos to deployment questions
Across the sector, the central challenge is no longer just whether humanoids can move dynamically, but whether they can work safely, cheaply, and continuously in real facilities. That means proving battery economics, human-safe behavior, certification readiness, and integration with warehouse and factory software. The companies making the strongest case are those pairing autonomy with measurable floor hours and manufacturable systems.
CONCLUSION
The humanoid race is increasingly defined by industrial credibility rather than viral stunts. Safety certification, factory-scale production, and reliable operation around people now appear to be the real thresholds for commercial adoption.
Ask a question