Engineering Modern Humanoid Locomotion: Kinematics, Dynamics & Actuation
The commercialization of full-scale humanoid robots—championed by Boston Dynamics' electric Atlas, Tesla Optimus, Figure 02, and Agility Digit—relies on high-frequency closed-loop balance. Transitioning from hydraulic power plants to high-density flux-switching electric actuators requires rigorous torque-budgeting to avoid thermal saturation during high-acceleration double-support and swing phases.
1. Inverse Kinematics (IK) & Foot Trajectory Generation
During bipedal walking, the center of mass (COM) describes a cycloidal 3D trajectory governed by inverted pendulum dynamics (LIPM: Linear Inverted Pendulum Model). The swing foot follows a quintic polynomial path to ensure zero-impact velocity at heel strike:
From Cartesian endpoints (xfoot, yfoot, zfoot), joint angles (θhip, θknee, θankle) are resolved via geometric inverse kinematics, adhering to link lengths L1 (femur) and L2 (tibia).
2. Zero-Moment Point (ZMP) & Dynamic Balance Criterion
Static center-of-gravity projection is insufficient for dynamic gaits where acceleration forces and inertia tensors generate tipping moments. Vukobratović’s Zero-Moment Point defines the instantaneous point on the ground where the net horizontal tipping moment equals zero:
If xzmp remains strictly inside the convex hull of the support foot (single-support phase) or both feet (double-support phase), the robot will not overturn. If xzmp reaches the boundary (heel or toe edge), the foot begins rolling, triggering an uncommanded fall unless rapid reactive stepping occurs.
3. Recursive Newton-Euler Joint Torque Formulations
Dynamic joint torques τ are calculated via generalized equations of motion:
Where M(q) is the mass-inertia matrix, C represents Coriolis and centrifugal terms, G is gravity loading, and JT Fext reflects ground reaction forces and carried payload masses. In high-speed bipedal locomotion, the knee torque peaks during early stance shock absorption, demanding over 250 N·m in typical 80 kg humanoids.
| Actuator Subsystem | Primary Joint Role | Typical Peak Torque | Gear Reduction Type |
|---|---|---|---|
| Hip Pitch / Roll | Torso stabilization, sagittal acceleration | 180 – 260 N·m | Planetary / Cycloidal (25:1 to 50:1) |
| Knee Flexion / Ext | Stance phase support, payload lifting | 240 – 380 N·m | Roller Screw / High-ratio Cycloidal |
| Ankle Pitch | Toe-off impulse, push-off velocity | 140 – 220 N·m | Bilateral Linear Actuators / 4-bar linkage |
| Shoulder / Elbow | Counter-swing inertia, payload carriage | 60 – 120 N·m | Quasi-Direct Drive (QDD) |
Frequently Asked Technical Questions
Why have electric humanoids largely superseded hydraulics in 2026?
While hydraulic actuators deliver extraordinary force density (as seen in early Boston Dynamics Atlas iterations), electric actuators offer higher energy efficiency (85%+ vs 30-45% for hydraulic pumps), zero risk of fluid contamination, silent operation, and direct motor current sensing for backdrivable admittance and impedance control.
How does carrying a heavy payload alter joint torque profiles?
Carrying a front-held box shifts the combined center of mass (COM) forward, forcing the hip pitch actuators to produce continuous extensor torque to prevent forward tipping. Simultaneously, stance-phase knee actuators face heightened moment arms, increasing RMS thermal heating and battery discharge rates.
What is the difference between ZMP and Capture Point (CP)?
The Zero-Moment Point evaluates whether the robot is currently stable over its active foot contact. The Capture Point (pioneered by Pratt et al.) predicts where the robot must place its next swing foot on the ground to come to a complete, balanced standstill.