Research Analyzer
← Back ICRA 2026

A Whole-Body Control Framework for Human-Like Walking with Knee Stretch on Flat-Foot Humanoids

Taehyun Kim,, Sookyoung Yoo,, Myo-Taeg Lim and Yonghwan Oh,∗

PDF

AI summary

Key figure (auto-extracted from paper)
A hierarchical control framework enables flat-foot humanoids to achieve natural, human-like heel-to-toe walking with significantly longer strides and reduced knee effort without requiring actuated toe joints.
Heel-to-toe walking Flat-foot humanoid Whole-body control Damped least squares IK Finite state machine Quadratic programming

Problem

Flat-foot humanoids struggle to replicate natural human gait and achieve long strides due to kinematic singularities at full knee extension and restrictive flat-foot walking patterns. Existing methods either require complex actuated toe joints or rely on rigid, non-adaptive trajectories.

Approach

The method uses a four-state finite state machine with mixed time and event-based transitions to guide a hierarchical controller. It combines a 5-DoF damped least squares inverse kinematics task for the swing leg, a shank-to-vertical angle-based ankle policy for adaptive heel strikes, and a real-time quadratic programming whole-body controller with phase-specific contact constraints.

Key results

  • Max step length reaches 72 cm (39.6% of robot height), comparable to human stride
  • Maintains stability while significantly reducing stance knee joint effort
  • Enables adaptive heel-to-toe rolling on flat-foot robots without actuated toe joints
  • Successfully resolves near-singular knee extension configurations via DLS-IK and relaxed contact constraints

Why it matters

Provides a practical, adaptive locomotion solution for the majority of existing flat-foot humanoids, bridging the gap between stable machine walking and natural human gait.

Abstract

Achieving a natural, human-like heel to toe gait on flat-foot humanoids is a challenging task, as their typical parallel foot walking pattern limits step length and appears unnatural. To address this, we present a hierarchical control framework guided by a four-state Finite State Machine with mixed time and event based transitions. The controller manages near-singular knee extension using a 5-DoF damped least squares inverse kinematics task for the swing leg. A key innova- tion is a late-swing ankle policy based on the shank-to-vertical angle, which allows a natural heel strike to emerge and adapt to walking speed. A real-time whole-body controller based on quadratic programming, incorporating the robot’s full- body dynamics, realizes these motions while satisfying contact constraints. In simulation, our method achieves significantly longer steps with maintained stability and reduced stance knee effort, reaching a maximum step length of 72 cm (39.6% of the robot’s height), which is comparable to that of a human.

Index terms

Human and Humanoid Motion Analysis and Synthesis Humanoid and Bipedal Locomotion Humanoid Robot Systems

Related papers