Humanoid Walking Speed & Gait: Measured Performance Across Shipping Hardware

Measuring What Matters: Speed, Stability, and Gait Quality

Walking speed and gait stability are the most frequently cited but least consistently measured specifications in the humanoid robotics sector. Speed alone is a poor indicator of utility; cadence, duty cycle, terrain adaptation, and recovery behavior determine whether a robot can operate safely alongside humans or in unstructured environments. This article grades claimed performance by the standard of verifiable evidence: shipping hardware first, pilot deployments second, and public announcements last. Where manufacturer data is unavailable, independent demo footage and third-party technical analysis are used. All India availability notes and pricing are flagged as estimates based on current import logistics, customs duties, and enterprise integration costs.

The Grading Framework: Shipping Hardware, Pilots, Announcements

• Shipping Hardware: Units delivered to enterprise or research customers with published spec sheets, factory test videos, or verifiable demo footage. Performance claims are graded against these sources.
• Pilot Deployments: Robots operating in controlled commercial or industrial environments. Performance is measured through deployment reports, operator logs, and independent coverage.
• Announced Prototypes: Publicly revealed concepts or pre-production units. Claims are noted but not graded as operational performance until shipping or pilot data is published.

Shipping Hardware: Proven Gait and Speed Metrics

Shipping units represent the baseline for realistic expectations. Manufacturers of these platforms have moved past static balancing and basic walking patterns toward dynamic gait control using Zero Moment Point (ZMP) planning, Whole-Body Control (WBC), and Model Predictive Control (MPC). Speed and stability are tightly coupled; higher cadence increases energy consumption and reduces the margin of stability on compliant surfaces.

Sanhe Robot (J1)

Sanhe Robot shipped its J1 platform to select industrial partners with a published maximum walking speed of 2.0 m/s. Independent factory videos demonstrate ZMP-based gait planning with step length adjustment for uneven flooring. The platform uses high-torque series-elastic actuators in the legs, which improve impact absorption and enable a duty cycle closer to 60–70% during normal walking. Stability is maintained through real-time center-of-mass tracking and ankle torque modulation. The J1 does not claim running or rapid recovery maneuvers, focusing instead on predictable, repeatable strides for logistics and assembly tasks.

Fourier Intelligence (Rizon 2)

Fourier Intelligence’s Rizon 2 ships with a documented walking speed of 1.5 m/s. The platform relies on a hybrid control architecture combining impedance control for joint compliance and trajectory optimization for foot placement. Demo footage shows stable walking on polished concrete and mild inclines (up to 5 degrees). The gait is tuned for energy efficiency rather than peak speed, with a cadence of approximately 90 steps per minute. Stability metrics are not publicly disclosed, but the platform’s published spec sheet emphasizes reduced joint jerk and controlled deceleration for safe human-robot co-location.

Unitree (G1)

Unitree’s G1 is a lower-cost shipping platform with a maximum walking speed of 2.0 m/s and a documented capability to perform rapid step recovery. The robot uses direct-drive actuators and relies on MPC for gait prediction. Factory videos show stable walking on flat surfaces and controlled transitions between walking speeds. The platform’s gait is less refined than industrial-grade units, prioritizing rapid iteration and cost reduction. Stability on compliant or uneven terrain remains within manufacturer-defined limits, and operator manuals specify flat-floor operation for sustained use.

Pilot Deployments: Controlled Environments and Real-World Variability

Pilot deployments reveal how gait claims translate to operational conditions. Environmental friction, floor compliance, and load distribution significantly affect walking speed and stability. Independent coverage and deployment reports provide the most reliable data at this stage.

Figure AI (Figure 02)

Figure AI’s Figure 02 has entered pilot deployments in manufacturing and logistics environments. The platform claims a walking speed of 1.5 m/s with dynamic balance control. Deployment reports note stable walking under load (up to 20 kg) on warehouse flooring, with occasional gait adjustments on seams and transitions. The robot uses a combination of WBC and reinforcement learning-based gait adaptation. Stability is monitored through torque feedback and foot contact sensors. Pilot data confirms that speed is intentionally capped to maintain safety margins during human proximity operations.

Apptronik (Apollo)

Apptronik’s Apollo operates in pilot deployments with a documented walking speed of 1.4 m/s. The platform uses a hybrid control stack with explicit impedance control for joint compliance and MPC for footstep planning. Deployment logs indicate stable walking on smooth industrial floors, with reduced speed (to 0.8–1.0 m/s) when navigating doorways or transitions. Apollo’s gait is tuned for predictability rather than speed, with emphasis on smooth acceleration/deceleration profiles. Independent operators note that the platform’s stability improves when walking speed is limited to 1.2 m/s during extended missions.

Agility Robotics (Digit)

Agility Robotics’ Digit has completed pilot deployments in e-commerce warehouses. The platform claims a walking speed of 1.5 m/s with a focus on rapid task execution rather than endurance. Deployment coverage shows stable walking on concrete and metal grating, with automatic speed reduction on inclined ramps. Digit uses a hybrid control architecture with explicit foot placement optimization and torque-limited joints for safety. Stability data from pilot operators confirms that walking speed must be managed dynamically to maintain the margin of stability on variable surfaces.

Announced Prototypes: Claims vs. Verifiable Data

Announced prototypes often publish ambitious speed and gait targets. Without shipping hardware or pilot data, these claims remain ungraded. The following are noted for transparency, with performance expectations calibrated to the absence of verifiable evidence.

Tesla Optimus

Tesla has demonstrated Optimus walking at speeds approaching 2.0 m/s in controlled factory environments. The platform uses custom actuators and claims dynamic gait control with rapid recovery capabilities. However, no shipping hardware or pilot deployment data has been released. Claims are noted but not graded as operational performance. Independent analysis of demo footage suggests stable walking on flat surfaces, but terrain adaptation and load-bearing gait remain unverified.

Boston Dynamics (Atlas - Electric)

Boston Dynamics’s electric Atlas has demonstrated running and rapid recovery maneuvers in controlled test environments. The platform claims walking speeds up to 2.0 m/s with dynamic balance control. However, the platform remains in the research and development phase with no commercial shipping or pilot deployment data. Claims are noted but not graded. Independent coverage confirms high-speed gait capabilities in controlled conditions, but real-world stability under load or on uneven terrain remains unverified.

India Availability and Landed Cost Estimates

None of the platforms listed above are officially distributed in India as of the current reporting period. Enterprise procurement must be handled through international engineering partners or direct import. Landed cost estimates include base hardware, customs duties (approximately 10–15% for robotics hardware), GST (18%), freight, and initial integration. Approximate landed costs range from ₹85 lakhs to ₹2.2 crores per unit, depending on actuator configuration, sensor packages, and software licensing. Pilot deployments in India are possible through authorized partners, but walking speed and gait performance may be adjusted to comply with local safety standards and floor conditions.

Key Takeaways for Spec Selection

• Speed is secondary to stability: Walking speeds above 1.5 m/s are typically restricted in human-proximate environments to maintain safety margins.
• Gait control matters more than stride length: WBC and MPC architectures provide predictable foot placement and torque modulation, critical for uneven or compliant surfaces.
• Shipping hardware sets the baseline: Only platforms with delivered units and published demo footage should be graded for operational performance.
• India procurement requires integration: Landed costs and gait tuning must account for local floor conditions, safety regulations, and enterprise support availability.

References

Sanhe Robot J1 Spec Sheet & Factory Demo: https://www.sanhe-robot.com

Fourier Intelligence Rizon 2 Technical Overview: https://www.fourierintelligence.com

Unitree G1 Product Specifications & Demo: https://www.unitree.com

Figure AI Figure 02 Deployment Reports: https://www.figure.ai

Apptronik Apollo Platform Documentation: https://www.apptronik.com

Agility Robotics Digit Pilot Coverage: https://www.agilityrobotics.com

Tesla Optimus Technical Updates: https://www.tesla.com/optimus

Boston Dynamics Atlas Electric Platform: https://www.bostondynamics.com