Adaptive Control of Lower-Limb Assistive Exoskeleton for Rehabilitation Using Deep Reinforcement Learning

Deep Reinforcement Learning-Based Control Strategies for Minimizing Interaction Forces in Lower-Limb Wearable Robots

Results demonstrate that DRL-based methods, particularly when combined with traditional controllers, improve both force reduction and motion stability over conventional control strategies.

Dynamics-Based Collaborative Control for an Exoskeleton-Walker System via Deterministic Learning

Lower limb exoskeletons and mobile robots hold great potential in improving motor function rehabilitation for patients with limb dysfunction. However, their widespread application is limited by the substantial time and effort investment required from professional rehabilitation therapists. A significant challenge in achieving autonomous and intelligent rehabilitation lies in addressing the coordinated control between the exoskeleton and the robotic walker. This paper proposes a novel collaborative learning control strategy based on deterministic learning, which aims to achieve high-performance coordinated control through precise closed-loop dynamics modeling of the exoskeleton-walker system. First, radial basis function neural networks (RBFNNs) are employed to approximate the system dynamics during the coordinated control process. Utilizing deterministic learning theory, it is rigorously demonstrated that, under persistent excitation conditions, the unknown dynamics of the system can be accurately approximated and stored as constant neural networks. Subsequently, an experience-based collaborative learning controller is designed, enabling autonomous coordinated control of the human-robot system and offering a viable approach for its broader application. The effectiveness and superior performance of the proposed control strategy are validated through experiments conducted on the CoppeliaSim robotic platform. Note to Practitioners—This work is intended for researchers and engineers working on rehabilitation robotics, particularly those focusing on lower limb exoskeletons and mobile robotic walkers. One of the main barriers to the practical deployment of these systems is the reliance on continuous assistance from rehabilitation professionals during use. To address this, we propose a deterministic learning-based collaborative control strategy that enables accurate modeling and reuse of system dynamics, ultimately achieving autonomous coordination between the exoskeleton and robotic walker. This reduces reliance on manual intervention and opens up the possibility for long-term, high-performance rehabilitation training in clinical and home environments. Practitioners can leverage this approach to develop smarter, more adaptive rehabilitation systems that respond to patient needs with greater precision and autonomy.

Performance Comparison of Classical and Robust Control Strategies for a Lower-Limb Rehabilitation Exoskeleton

Lower-limb rehabilitation exoskeletons have emerged as a promising complementary technology to conventional therapy, enabling repetitive, intensive, and personalized gait training. However, achieving accurate trajectory tracking while maintaining robustness against parametric uncertainties, external disturbances, and unpredictable human–robot interaction remains a significant control challenge due to the highly nonlinear dynamics of coupled human–exoskeleton systems. This paper presents an experimental performance comparison of five control strategies for gait rehabilitation exoskeletons, including a classical proportional–integral–derivative (PID) controller, a model-based proportional–derivative controller with gravity compensation (PD+G), a computed torque sliding mode controller (CT-SMC), a computed torque–super-twisting sliding mode controller (CT–ST-SMC) and a hybrid backstepping–super-twisting sliding mode controller (BS–ST-SMC). All the controllers were implemented on the same lower-limb rehabilitation exoskeleton under identical operating conditions. The experimental results demonstrate that the proposed BS–ST-SMC architecture outperforms classical and traditional robust approaches, particularly in mitigating chattering and managing human–robot interaction uncertainties. Specifically, the BS–ST-SMC achieved the highest tracking precision with a mean squared position error (MSEp) of 1.32×10−3rad2 and effectively synchronized with the user by reducing the phase lag to just 4.22° at the knee joint. Their overall performance was evaluated using the following metrics: mean squared position error (MSEP), mean squared velocity error (MSEv), peak error, phase lag, jerk index, peak torque, and peak power.

Simulation study of exoskeleton-musculoskeletal coupling for exoskeleton assistance via multi-agent reinforcement learning.

Dynamic Parameter Identification of a Lower-Limb Exoskeleton Using RLS–AGWO

Accurate dynamic parameters are required for model-based control of lower-limb exoskeletons, but limited excitation, transmission friction, and assembly-dependent uncertainty can degrade conventional estimates. This study examines a two-stage method that combines recursive least squares (RLS) with an adaptive grey wolf optimizer (AGWO). Offline RLS tracks the base-parameter trajectory and expands its post-convergence extrema to construct a finite search space; a non-smooth friction severity index then modulates the GWO convergence schedule. The method was evaluated on a pedestal-mounted, single-degree-of-freedom hip mechanism using a 5 s calibration trajectory and a separate 7 s validation trajectory. Deterministic least squares (LS) and bound-constrained least squares (BCLS) were compared with standard PSO, RLS–PSO, RLS–GA, RLS–GWO, and RLS–AGWO. Each stochastic method used a population of 30, with 80 iterations (2400 fitness evaluations) and 30 independent seeds. On the independent trajectory, BCLS obtained an RMSE of 0.1152 Nm. Median validation RMSEs were 0.1152, 0.1152, 0.1562, and 0.1516 Nm for RLS–PSO, RLS–GA, RLS–GWO, and RLS–AGWO, respectively. Thus, the adaptive schedule improved median GWO error by 3.0%, but deterministic BCLS was both more accurate and faster for the present linear-in-parameters model. AGWO is therefore not mathematically necessary for the current convex objective; its potential advantage should be tested with genuinely nonlinear friction parameterizations. The conclusions remain limited to a single-axis pedestal experiment and do not establish performance during human-worn gait.

A Composite Variable Impedance Control Architecture with Adaptive Feedforward Assistance for Rehabilitation Exoskeletons

Home-based rehabilitation exoskeletons often suffer from control instability due to low-cost force sensors. This paper presents a robust, sensorless Composite Variable Impedance Control architecture that separates trajectory tracking (virtual stiffness K) from active assistance (adaptive feedforward torque τassist). By eliminating high-frequency force feedback, the system ensures intrinsic stability. Experiments on the CURE platform demonstrate independent modulation of compliance (RMSE 2.64° to 15.90°) and effective assistance during simulated weakness, reducing tracking RMSE from 13.77° to 3.12°. Results show τassist contributes 59.6% of total torque, enabling "High-Assistance, High-Compliance" interaction without reactive stiffening. This provides a stable execution layer for advanced, bio-signal-driven "Assist-as-Needed" (AAN) therapies.