Matt Strong

Matt Strong

CS PhD at Stanford University

← Paper Digest

Feedback Control of an Exoskeleton for Paraplegics: Toward Robustly Stable Hands-free Dynamic Walking

Omar Harib, Ayonga Hereid, Ayush Agrawal, Thomas Gurriet, Sylvain Finet, Guilhem Boeris, Alexis Duburcq, M. Eva Mungai, Matthieu Masselin, Aaron D. Ames, Koushil Sreenath, Jessy Grizzle

Seminal February 2018

TL;DR The authors port hybrid-zero-dynamics control from bipedal robots to the 12-actuator ATALANTE exoskeleton so complete paraplegics walk without crutches. Three patients walked about 9-10.5 m hands-free at 0.11-0.15 m/s, and a newer learned controller recovers from 750 N pushes in simulation where a fixed gait fails above 300 N.

What they do

Exoskeletons such as EksoGT, ReWalk and HAL need crutches for balance; REX is hands-free but walks statically at about 0.05 m/s. This paper brings bipedal-robot control (virtual constraints, hybrid zero dynamics) to ATALANTE, a 12-actuator exoskeleton from Wandercraft, so paraplegic users walk dynamically without crutches. Patients tested a single-gait controller (PHZD); a newer one combining trajectory optimization and supervised learning (G-HZD) is tested in simulation.

How it works

  • Model. User and exoskeleton form one rigid 18-DOF floating-base model walking flat-footed, with 0.7 s steps, a 27 cm stance width and constraints on ZMP, friction, foot impact velocity and torso lean.
  • PHZD. Virtual constraints on forward hip velocity and joint angles are driven to zero by input-output linearization. Direct collocation picks their parameters so that one periodic gait is stable and invariant through foot impact.
  • G-HZD. Periodic gaits over a range of hip velocities, plus three-step transition gaits converging to a target gait, are regressed into a function of hip velocity and time. It re-plans joint targets right after a disturbance; feedback linearization with PD gains tracks them.

Results

  • Three complete paraplegics (lesions T6-T12) walked 8.9, 10.56 and 9.5 m at 0.11-0.15 m/s with no outside assistance.
  • In Gazebo, the learned controller recovers from hip pushes of 750 N forward and 650 N backward; a fixed-gait controller fails above 300 N and 100 N. It recovers by changing step length.
  • It tracks commanded speeds up to 0.15 m/s with a small steady-state error, withstands a constant ±75 N torso force for 10 s, and 100 Nm knee torques (modeled spasticity) raise knee tracking error from 0.0130 to at most 0.0180 rad.

Limitations

The learned controller has not run on hardware. Patient walking was slow, straight-line only, with a gait chosen and tuned per patient; failed trials came from foot scuffing or loss of lateral balance. The tests are not a clinical evaluation.

Figures

From Figure 2: A user in ATALANTE and the kinematic diagram of the lumped human-exoskeleton system. The user is strap...
From Figure 2: A user in ATALANTE and the kinematic diagram of the lumped human-exoskeleton system. The user is strapped in at the ankle, shin, thigh, abdomen and torso; link lengths in green are adjustable.
From Figure 4: Controller structure. The Guard Checker decides when to swap legs, the Control Policy specifies the de...
From Figure 4: Controller structure. The Guard Checker decides when to swap legs, the Control Policy specifies the desired gait, and the Low-level Controller turns it into joint trajectories tracked by PD control.
From Figure 17: Forward hip velocity in simulation after impulsive pushes from -650 N to +750 N with the supervised-l...
From Figure 17: Forward hip velocity in simulation after impulsive pushes from -650 N to +750 N with the supervised-learning controller.
From Figure 17: The same test with a fixed-gait controller, shown for pushes from -100 N to +300 N. Both controllers ...
From Figure 17: The same test with a fixed-gait controller, shown for pushes from -100 N to +300 N. Both controllers failed under larger pushes than those shown.
Abstract

This manuscript presents control of a high-DOF fully actuated lower-limb exoskeleton for paraplegic individuals. The key novelty is the ability for the user to walk without the use of crutches or other external means of stabilization. We harness the power of modern optimization techniques and supervised machine learning to develop a smooth feedback control policy that provides robust velocity regulation and perturbation rejection. Preliminary evaluation of the stability and robustness of the proposed approach is demonstrated through the Gazebo simulation environment. In addition, preliminary experimental results with (complete) paraplegic individuals are included for the previous version of the controller.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.