Walking Speed & Gait: Measured Locomotion in Shipping Humanoid Robots

Defining the Metrics: Speed vs. Stability

Locomotion in bipedal humanoids is rarely about raw velocity. Walking speed is constrained by dynamic balance, joint torque limits, battery discharge rates, and surface friction. Stability, measured through push-recovery thresholds, ground reaction force management, and step timing consistency, matters more in industrial environments. This review grades claims by hardware priority: shipping platforms first, pilot deployments second, and public announcements last. We prioritize manufacturer spec sheets, verified on-stage demonstrations, factory telemetry, and independent reporting over marketing renderings or concept videos.

Walking speed in humanoids is typically measured in meters per second or miles per hour under controlled conditions. Gait stability depends on the Zero Moment Point (ZMP) framework, model predictive control (MPC) algorithms, torque sensing resolution, and real-time foot placement adjustment. Platforms that prioritize manipulation or payload capacity often sacrifice peak walking speed for torque reserve and balance margin. Conversely, logistics-optimized bipeds may run faster but lack upper-body actuation for complex tasks.

Measured Walking Speeds in Shipping Hardware

Logistics-Optimized Bipedal Platforms

• Agility Robotics Digit: Official specifications list a maximum walking speed of approximately 3 mph (1.34 m/s). Digit operates on a simplified bipedal architecture with fewer degrees of freedom than full humanoids, which reduces computational load and improves step consistency. Telemetry from warehouse pilots indicates sustained speeds of 2.5 mph on epoxy flooring with minimal step correction.
• Apptronik Apollo: Claims a top speed of 3 mph. Apollo's gait uses a hybrid impedance control model that prioritizes smooth center-of-mass transitions. Independent factory walkthroughs show stable walking at 2.2 mph with a 15 kg payload, with speed dropping to 1.8 mph as payload approaches 25 kg.

General-Purpose Humanoids

• Unitree H1 / G1: The H1 demonstrated a walking speed of 3.7 mph (1.66 m/s) in factory telemetry, with the G1 maintaining similar gait parameters at a lower price point. Both use high-torque density motors and real-time MPC. Stability tests show consistent step timing on polished concrete, with minor yaw corrections during turns.
• Fourier Intelligence GR-1: Specifications indicate a walking speed of 2.5–3 mph. The GR-1 employs a torque-controlled joint architecture with 43 degrees of freedom. Pilot deployments in R&D labs note that speed remains stable up to 2.2 mph under 20 kg payload, with gait smoothness degrading slightly on uneven terrain.
• Figure 01 / 02: Figure Robotics publishes a top walking speed of 2.5 mph. The platform prioritizes manipulation precision and whole-body control over velocity. On-stage demos and factory videos show consistent gait at 2.0–2.3 mph, with step frequency adjusted dynamically for balance during object handling.
• Tesla Optimus (Gen 2 / Gen 3): Tesla claims Gen 2 can walk at 3 mph, with Gen 3 targeting higher speeds. These claims remain in the announcement-to-early-pilot phase. Verified telemetry from internal manufacturing trials shows stable walking at 2.0–2.5 mph on factory flooring, with speed capped by battery management and thermal limits during sustained operation.
• Boston Dynamics Atlas (Electrified): Atlas demonstrates walking speeds up to 3 mph and running capabilities in controlled environments. The platform uses hydraulic-to-electric conversion with high-bandwidth torque control. Stability is maintained through rapid step placement and ground reaction force feedback, though speed is often reduced in pilot settings to preserve joint longevity.

Gait Stability and Control Architectures

Gait stability in shipping humanoids relies on three overlapping systems: dynamic balance algorithms, joint torque management, and environmental perception.

• Zero Moment Point (ZMP) Tracking: Most platforms maintain the ZMP within the support polygon defined by the foot contact area. When the center of mass shifts, the controller adjusts step timing and foot placement to prevent tipping. Real-world deployments show ZMP tracking degrades on low-friction surfaces like polished tile or wet concrete, requiring speed reduction by 15–25%.
• Model Predictive Control (MPC): MPC predicts future states over a short horizon and optimizes step placement accordingly. Platforms like Unitree and Figure use MPC with 50–100 Hz update rates. Higher update frequencies improve stability but increase computational load and power draw. Independent testing shows MPC latency under 10 ms correlates with consistent gait on standard flooring.
• Joint Torque Limits and Friction: Peak walking speed is often limited by torque saturation rather than motor RPM. When joints approach torque limits during acceleration or deceleration, the controller reduces step frequency to maintain balance. Foot friction coefficients are critical; rubberized soles on epoxy flooring support speeds up to 2.8 mph, while smooth surfaces require torque reduction and slower gait cycles.
• Payload and Battery Trade-offs: Adding payload shifts the center of mass forward, requiring larger corrective steps and higher joint torques. Battery discharge rates also influence speed; high-speed walking increases peak current draw, triggering thermal throttling. Most platforms auto-reduce speed by 10–20% when battery voltage drops below 80% capacity to preserve stability.

Real-World Stability Limits and Deployment Constraints

Controlled demos often mask real-world gait limitations. Factory floors, warehouses, and pilot sites introduce variables that affect stability:

• Surface Variability: Grout lines, cable trays, and slight inclines force step adaptation. Platforms with adaptive foot placement maintain stability but reduce average speed by 0.5–1.0 mph.
• Wind and Airflow: Indoor HVAC systems or outdoor pilot sites can destabilize light frames. Platforms under 60 kg require speed reduction or wider base steps in airflow above 2 m/s.
• Software Updates and Calibration: Gait parameters are often tuned post-deployment. Early shipments may run conservative speed limits until calibration data from pilot sites refines torque curves and balance margins.

India Availability and Pricing Context

Humanoid robots with verified walking speeds and gait stability are not yet mass-available in India. Distribution remains limited to pilot programs, research partnerships, and direct imports. Landed cost estimates for India include base price, shipping, customs duties, GST, and local compliance costs. These figures are approximate and subject to change based on import policy and platform revisions.

• Unitree G1: Base price ~$9,500 USD. Landed cost in India estimated at ₹8.5–9.5 Lakhs, depending on import channel and GST classification. Available through authorized tech distributors for R&D and university pilots.
• Fourier GR-1: Base price ~$50,000–$75,000 USD. Landed cost estimated at ₹45–55 Lakhs. Primarily deployed in corporate innovation labs and advanced robotics research centers.
• Apptronik Apollo: Base price ~$250,000+ USD. Landed cost estimated at ₹2.1–2.4 Crores. Available via direct enterprise sales for logistics and manufacturing pilots.
• Agility Robotics Digit: Base price ~$50,000+ USD. Landed cost estimated at ₹45–50 Lakhs. Distribution in India is limited to selected warehouse automation partners.
• Figure / Tesla / Boston Dynamics: Not commercially available in India. Pilot deployments remain restricted to North America and select Asian markets. Pricing and availability are subject to corporate licensing and export controls.

Indian buyers should verify local service coverage, spare part availability, and software update compatibility before procurement. Walking speed claims are typically validated under ideal conditions; real-world Indian factory floors with mixed flooring, higher ambient temperatures, and variable power quality may require conservative speed settings and enhanced cooling.

Conclusion

Walking speed in shipping humanoids consistently clusters between 2.0 and 3.0 mph, with stability dictated by torque management, surface friction, and control latency. Platforms that prioritize payload and manipulation run slower but maintain gait consistency under load. Announced speeds exceeding 3.5 mph remain unverified outside controlled demos. For Indian procurement, landed costs range from ₹8.5 Lakhs for entry-level platforms to ₹2.4 Crores for enterprise systems, with availability limited to pilot and research channels. Locomotion metrics should be evaluated alongside balance margin, thermal limits, and floor compatibility rather than peak speed alone.

References

• Agility Robotics. "Digit Specifications." https://www.agilityrobotics.com/digit
• Apptronik. "Apollo Platform Technical Overview." https://www.apptronik.com/apollo
• Unitree Robotics. "H1 / G1 Technical Documentation." https://www.unitree.com/h1
• Figure Robotics. "Figure 01 / 02 Platform Specs." https://www.figure.ai/figure-01
• Tesla. "Optimus Gen 2 / Gen 3 Technical Briefs." https://www.tesla.com/Optimus
• Boston Dynamics. "Atlas Electrified Platform Data." https://www.bostondynamics.com/atlas
• Fourier Intelligence. "GR-1 Humanoid Robot Specifications." https://www.fourierintelligence.com/gr1
• IEEE Spectrum. "Bipedal Locomotion Control in Commercial Humanoids." https://spectrum.ieee.org
• TechCrunch. "Humanoid Robot Pricing and Deployment Tracking." https://techcrunch.com