Digit, a humanoid robot developed by Agility Robotics, stood out at Web Summit 2024 with its distinctive teal frame, reverse-jointed legs, and swappable arms. Standing at 1.75 meters and weighing 72 kilograms, it is engineered for environments built for humans, with feet rather than wheels to navigate stairs and uneven terrain. Its flat, spatula-like hands are optimized for gripping and moving boxes, while the reverse leg design prevents accidental kicks when bending to lift objects.
Digit operates fully autonomously, without teleoperation, and is trained on multiple commercial large language models (LLMs) to adapt to varied workflows. Agility has already deployed fleets in GXO’s warehouses, with early tasks including moving tote bags in a Connecticut Spanx facility. This arrangement marked what Agility described as the first Robots-as-a-Service (RaaS) deployment of humanoids. A second deployment is planned at the Schaeffler Group, which aims to integrate a significant number of units across its 100 plants by 2030. Schaeffler has also taken a minority stake in Agility.
CEO Peggy Johnson emphasized the role of humanoids in alleviating physically taxing, repetitive labor. “It will make life easier for human workers,” she said. “What we’re focussed on is augmenting humans… it can throw out knees and it’s dirty, repetitive mind-numbing work.” She noted that in the U.S., logistics has roughly one million unfilled positions, and robots could allow humans to shift toward supervisory and management tasks.
Hardware development has been a decade-long effort for Agility, with three years of in-facility trials refining use cases and mechanical design. Advances in processing power now allow Digit to leverage diverse consumer-grade LLMs for semantic understanding and task flexibility. “We take simulation data to teach Digit new skills,” Johnson explained. “With that AI and LLM we can start to improve Digit’s semantic intelligence… The robot can be doing one job in the morning and another in the afternoon, all supported by AI.”
Digit’s behavior can vary depending on the LLM. Johnson described how an Anthropic-trained unit might signal affection with digital heart eyes, while another model might arrange boxes into a heart shape. The robot’s perception suite includes one LiDAR sensor and seven cameras positioned in its neck and waist, with downward-facing units tracking position and forward-facing ones identifying objects. Acoustic sensors enable voice commands, though in noisy factory settings most inputs are delivered via tablet.
Battery endurance remains a constraint. Current performance yields a 4:1 work-to-charge ratio—four minutes of activity per minute of charging—with a target of 10:1. Digit can autonomously dock for recharging when needed. Safety protocols require maintaining distance from humans due to limitations in fine motor control and complex interaction handling, though Agility is developing cooperative safety features for closer collaboration, with a demo expected mid-2025 and commercial rollout within 18–24 months.
On stage at Web Summit, Digit demonstrated laundry sorting. It successfully identified and placed a grey shirt into a basket, but struggled with unpracticed tasks, such as stacking a green shirt atop a grey one. Processing delays of around ten seconds occurred, and prompt specificity proved critical—”striped shirt” yielded better recognition than “green shirt.” When tasked with collecting three shirts, Digit opted for sequential handling rather than a single scoop, dropping one and requiring additional processing before completion.
Johnson acknowledged that scaling efficiency will depend on both improved battery capacity and a broader range of swappable hands. “You can think of the hands as being a tool for whatever it is you need… We’ll aim to match the right tool with the right use case,” she said. For industrial robotics, the interplay between adaptable hardware and AI-driven workflow intelligence is shaping a new phase in human-machine collaboration.