Multi-Contact Whole-Body Force Control for Position-Controlled Robots

TL;DR

The paper addresses the challenge of regulating contact forces on position-controlled robots during multi-contact tasks by modeling whole-body flexibility and deriving a derivative-based linearization of the quasi-static equilibrium. It introduces SEIKO, a two-stage, SQP-based pipeline consisting of SEIKO Retargeting (feasible posture and contact-wrench targets) and SEIKO Controller (QP to compute flexible-state corrections and commanded joint positions), enabling indirect control of contact forces via joint commands. The authors demonstrate real-time operation on the Talos humanoid at 500 Hz, with pushing, far-reaching, stair-climbing, and sloped-surface tasks, and show robustness to model errors through deliberate mass/load perturbations and contacting-mode transitions. This framework broadens the practical deployment of multi-contact strategies on existing position-controlled platforms by leveraging joint flexibility to achieve stable, safe, and aggressive contact behaviors.

Abstract

Many humanoid and multi-legged robots are controlled in positions rather than in torques, which prevents direct control of contact forces, and hampers their ability to create multiple contacts to enhance their balance, such as placing a hand on a wall or a handrail. This letter introduces the SEIKO (Sequential Equilibrium Inverse Kinematic Optimization) pipeline, and proposes a unified formulation that exploits an explicit model of flexibility to indirectly control contact forces on traditional position-controlled robots. SEIKO formulates whole-body retargeting from Cartesian commands and admittance control using two quadratic programs solved in real-time. Our pipeline is validated with experiments on the real, full-scale humanoid robot Talos in various multi-contact scenarios, including pushing tasks, far-reaching tasks, stair climbing, and stepping on sloped surfaces. Code and videos are available at: https://hucebot.github.io/seiko_controller_website/

Figures (9)

• Figure 1: Overview of our control pipeline (top), and illustrations of teleoperated multi-contact experiments on Talos humanoid robot (bottom).
• Figure 2: Control architecture for position-controlled robots: Operator's Cartesian commands are retargeted into a feasible whole-body configuration. The controller uses a joint flexibility model to adjust actuator position commands for contact wrench control and prevent exceeding joint torque limits.
• Figure 3: Force distribution tracking during pushing tasks. The Talos robot (left) pushes a vertical wall using its left hand, following a predefined hand force target trajectory. Plots display the desired and measured normal force for the left hand (top) and the sagittal tangential force for the left foot (bottom); comparing with control enabled (5 trials) and without (5 trials).
• Figure 4: Comparison of contact switch trials with and without SEIKO Controller. Initially, both feet and right hand are in contact. The operator teleoperated the robot to disable the right foot contact, lift the foot, and re-establish contact. Vertical contact forces $\bm{\lambda}^{\text{d}}, \bm{\lambda}^{\text{read}}$ (top row) and the desired vertical position of the right foot $\bm{X}^{\text{d}}_{\text{right foot}}$ (bottom row) are displayed.
• Figure 5: Impact of damping gain $K_d$ on Talos's torso oscillations. Short pushes are applied (left), and IMU's gyroscope measures sagittal plane oscillation for varying $K_d$ values (middle). Damping effect is quantified using logarithmic decrement metric from oscillation peaks (right).