Digit (original platform)
About
About Digit (original platform)
The original Digit is Agility Robotics' headless bipedal platform introduced in February 2019 and offered commercially in 2020. It extends the Cassie leg design with a torso, four-degree-of-freedom arms, perception sensors and onboard computers. Its arms support balance and recovery as well as carrying objects, while articulated legs provide mobility through spaces built for people.
Early work explored package handling and vehicle-to-door delivery with Ford, which became the first commercial customer. Agility refined the feet, sensors and computing hardware before the 2020 launch and reported commercial shipments from July that year. The torso contains two multi-core processors and a modular bay for additional computing. This platform preceded the March 2023 Digit redesign, which introduced a head, eyes and hands optimised for logistics totes.
Capabilities
Source-backed capabilities
- Multi-terrain locomotion
Research demonstration - 2025-04-14 - Hardware experiments reached 1.5 m/s forward walking and tested slippery surfaces, slopes, uneven ground and sand.
Source - Multi-terrain locomotion
Research demonstration - 2024-10-04 - The controller uses proprioceptive history without vision, after flat-ground pre-training and reinforcement-learning refinement on uneven terrain.
Source
Research
- No More Marching: Learning Humanoid Locomotion for Short-Range SE(2) Targets
2025-08-16 - Research Publication - Humanoids operating in real-world workspaces must frequently execute task-driven, short-range movements to SE(2) target poses. To be practical, these transitions must be fast, robust, and energy efficient.
Source - PPF paper improves Digit locomotion with model-assumption-based regularization
2025-04-14 - Research Publication - This model covers PPF paper improves Digit locomotion with model-assumption-based regularization for Digit in Reinforcement learning.
Source - Challenging-terrain locomotion paper trains Digit for real-world outdoor walking
2024-10-04 - Research Publication - This paper covers Challenging-terrain locomotion paper trains Digit for real-world outdoor walking for Digit in Reinforcement learning.
Source - Digit system-identification paper improves constrained-robot model tracking
2024-08-16 - Research Publication - Identifying the parameters of robotic systems, such as motor inertia or joint friction, is critical to satisfactory controller synthesis, model analysis, and observer design.
Source - Next-token prediction paper demonstrates real-world humanoid locomotion on Digit
2024-02-29 - Research Publication - This paper covers Next-token prediction paper demonstrates real-world humanoid locomotion on Digit for Digit in Reinforcement learning.
Source - Safe whole-body task-space control paper validates hardware experiments on Digit
2023-10-12 - Research Publication - Complex robotic systems require whole-body controllers to deal with contact interactions, handle closed kinematic chains, and track task-space control objectives.
Source - Digit paper learns robust bipedal walking with reinforcement learning
2021-03-29 - Research Publication - This paper covers Digit paper learns robust bipedal walking with reinforcement learning for Digit in Reinforcement learning.
Source
Timeline
- Agility and Oregon State publish GoTo controller for short-range Digit walking
2025-08-16 - Research Publication - A constellation-based reward couples translation and rotation, with benchmarks measuring energy, time to target and footstep count. Physical Digit v3 tests covered 72 short-range navigation tasks; the study reported a 90.23% success rate and excluded failed trials from its efficiency analysis.
Source - PreCi study demonstrates original Digit locomotion on challenging terrain
2025-04-14 - Research Publication - Hardware experiments reached 1.5 m/s forward walking and tested slippery surfaces, slopes, uneven ground and sand. Model-assumption-based regularization preserves useful model-based behavior while allowing the learned policy to adapt to harder conditions.
Source - HT-2 study demonstrates original Digit walking over challenging terrain
2024-10-04 - Research Publication - A single controller traversed more than four miles across five Berkeley-area hikes and steep streets in San Francisco. The controller uses proprioceptive history without vision, after flat-ground pre-training and reinforcement-learning refinement on uneven terrain.
Source - System-identification study improves original Digit model tracking
2024-08-16 - Research Publication - Iterative least squares estimates motor inertia and joint friction from measured data for robots with closed kinematic chains. Physical tracking tests compared the identified parameters with manufacturer defaults; the team released its implementation as ConstrainedSysID.
Source - Next-token study transfers locomotion control to original Digit hardware
2024-02-29 - Research Publication - The model used simulated trajectories from learned and model-based controllers, plus retargeted motion-capture and human-video data. A model trained on 27 hours of walking data transferred to physical Digit without real-world tuning and generalized to unseen commands including backward walking.
Source - Safe whole-body controller study validates original Digit hardware
2023-10-12 - Research Publication - The controller coordinates body, torso and arm tasks while constraining joint motion, balance and contact forces. Simulation and hardware experiments covered squatting, bowing, arm-height limits, walking and collision-avoidance tasks on the 45 kg, 20-motor platform.
Source - Reinforcement-learning study transfers robust walking to original Digit
2021-03-29 - Research Publication - The framework uses reduced policy inputs and outputs, with hardware-feasible starting poses to support transfer from simulation. Hardware tests maintained walking under external pushes and on terrain absent from training, without domain randomization or curriculum learning.
Source - Agility Robotics launches original Digit for commercial sale
2020-01-06 - Product Launch - Ford and Agility planned further research into indoor logistics and last-50-feet delivery, including communication between vehicles and the robot. The commercial design added upgraded feet, navigation sensors and customer-ready onboard computers after prototype testing in 2019. Two pre-production prototypes were scheduled for Ford’s CES booth in Las Vegas, January 7–10, 2020.
Source - Agility Robotics partners with Ford on last-mile delivery research
2019-05-22 - Partnership - The concept assigns longer road journeys to vehicles and sidewalks, grass and stairs to a walking robot. Agility would draw on Ford’s mapping and object-recognition research; a delivery test program using commercial Digit robots was planned for early 2020.
Source - Agility Robotics unveils original Digit bipedal robot
2019-02-19 - Product Launch - Four-degree-of-freedom arms are designed for object handling, balance and recovery; the company described tests carrying boxes up to 18 kg. Pricing was to be announced later in 2019, with deliveries planned for the first quarter of 2020; the prototype remained in testing.
Source