Agility Robotics Digit Patent Sharpens Humanoid Safety for Factories

EDITORIAL / OPINION

Agility Robotics secured U.S. Patent 12,560,948 on February 24, 2026, for a multi-level safety architecture on Digit that escalates responses based on real-time risk assessment. The system monitors nearby humans via sensors and machine learning, then slows motion, lowers the robot's posture to reduce fall impact, or executes a full stop requiring manual reset. This filing arrives as Digit moves from pilot programs into scaled logistics and manufacturing roles.

Patent Mechanics in Context

The patent, titled “Escalating hazard-response of dynamically stable mobile robot in a collaborative environment and related technology,” was filed February 28, 2025. It targets dynamically stable bipeds operating in shared spaces where fixed cages are impractical. Unlike static safety zones, the approach continuously evaluates proximity, velocity, and predicted trajectories to avoid unnecessary halts while meeting industrial standards.

Agility has already layered additional hardware protections onto Digit, including Category 1 stops that maintain actuator power during deceleration, a Safety PLC rated to Performance Level d, on-robot E-stops, and Functional Safety over EtherCAT. These elements complement the patented software layer rather than replace it. The combination addresses both regulatory expectations and insurer concerns that have slowed broader humanoid adoption.

Deployment Reality Check

Digit currently operates at partner sites including Toyota facilities and GXO logistics operations. Battery endurance reaches four hours with a 35-pound payload capacity, supporting continuous shifts in material handling. Safety certification milestones, such as passing NRTL field inspections recognized by OSHA, provide external validation beyond marketing claims.

Competitors face similar hurdles. Tesla Optimus Gen 2 and Figure 02 both emphasize collaborative workspaces yet have released fewer public details on graduated response protocols. Agility's early commercial footprints give it measurable data on uptime versus stop frequency that theoretical models cannot replicate.

Regulatory and Insurance Implications

Insurance carriers and OSHA inspectors increasingly require documented risk mitigation for mobile humanoids. A patent that explicitly describes escalating interventions offers a paper trail for audits. Facilities deploying fleets of 50 or more units will face questions about aggregate stop rates and recovery times; the system aims to minimize both by intervening proportionally.

Lowering posture as an intermediate step reduces kinetic energy in potential collisions, a detail that could influence liability calculations. Full stops remain the final resort, preserving operational continuity compared with designs that default to immediate shutdowns.

Competitive Landscape

Boston Dynamics Atlas (electric version) and 1X NEO pursue overlapping markets but rely more heavily on simulation-derived safety envelopes. Agility's patent centers on runtime adaptation using onboard perception, which may prove more robust in unstructured warehouse aisles where worker paths vary daily.

Unitree and Fourier models target lower price points yet lack equivalent public IP around dynamic risk escalation. The gap matters when scaling from single-unit pilots to multi-shift operations where cumulative downtime directly affects ROI.

Path to Fleet Scaling

Digit V5 deployments will test whether the patented logic maintains acceptable intervention rates under higher traffic. Partners report interest in expanding beyond initial tote-moving tasks, but sustained uptime above 90 percent will require the safety stack to evolve alongside task complexity.

Agility's Fremont operations and stated IPO timeline hinge on proving these mechanisms at volume. Concrete engineering filings like this one separate companies that can pass factory audits from those still iterating in simulation.

TECHNICAL BREAKDOWN

Architecture

The core architecture fuses multi-modal sensor input with machine-learning models that classify risk levels in real time. Escalation logic triggers actuator commands without requiring cloud round-trips, supporting low-latency decisions on the edge.

Actuators and Sensors

Digit integrates joint-level torque sensing, lidar, and camera arrays feeding the risk-assessment pipeline. Responses modulate gait parameters and center-of-mass height rather than relying solely on braking.

Limitations

The patent does not disclose exact sensor fusion weights or training datasets, leaving reproducibility questions for third-party auditors. Edge cases involving fast-moving humans or cluttered sightlines remain unquantified in public materials.

Unresolved Questions

How frequently do intermediate posture adjustments occur versus full stops in live deployments? What is the measured impact on cycle time when fleets exceed ten units in a single aisle? Future updates may address these metrics as more sites come online.