Results demonstrate that DRL-based methods, particularly when combined with traditional controllers, improve both force reduction and motion stability over conventional control strategies.
Abstract
Wearable robots, particularly lower-limb exoskeletons, have gained attention for their use in rehabilitation and mobility assistance. A key challenge in their design is achieving natural, low-stress human-robot interaction with minimal discomfort. Zero-force control has emerged as an effective strategy for reducing contact forces and enhancing comfort. In this study a deep reinforcement learning (DRL)-based control method for a two-degree-of-freedom lower-limb wearable robot was developed to track natural walking motions while minimizing interaction forces and adapting to varying user characteristics. A simulation environment was created using MuJoCo, and the DRL agent was trained using the TD3 algorithm. Three control approaches were tested: direct velocity setpoint control, adaptive proportional-derivative controller tuning, and adaptive lead compensator tuning. Experimental results showed that the adaptive methods reduced average interaction forces from 3.1 N to 1.7 N (thigh) and from 7.2 N to 2.7 N (shank). The maximum forces during the walking experiments were also lower for the adaptive methods, with the third method achieving a maximum of 15 N. These results demonstrate that DRL-based methods, particularly when combined with traditional controllers, improve both force reduction and motion stability over conventional control strategies.
Lower-limb rehabilitation exoskeletons have the potential to improve gait recovery after stroke by providing intensive and repetitive training. However, conventional control strategies often rely on fixed control parameters and exhibit limited adaptability to patient-specific characteristics, sensor noise, and dynamic uncertainties. This paper proposes an adaptive control framework that combines deep reinforcement learning (RL) with model-based impedance control for personalised lower-limb exoskeleton assistance. Patient-specific biological parameters are incorporated into the simulation environment and reward formulation to improve adaptability and robustness. Three state-of-the-art deep RL algorithms, Deep Deterministic Policy Gradient (DDPG), Twin Delayed Deep Deterministic Policy Gradient (TD3), and Soft Actor-Critic (SAC), are evaluated in a continuous control environment under varying signal-to-noise ratio (SNR) conditions ranging from 5 dB to noise-free conditions. Results demonstrate that TD3 achieves the most stable learning performance, obtaining a mean reward of −354.24 under noise-free conditions, while DDPG provides the highest joint-angle tracking accuracy with an RMSE of 0.0369 rad. SAC exhibits superior robustness in noisy environments, achieving the highest learning ratio of 0.51 at 5 dB SNR. Furthermore, the proposed personalised framework reduces tracking errors by up to 27% compared with non-personalised baseline approaches. The findings indicate that integrating patient-specific information with RL-based adaptive control can significantly enhance robustness, tracking performance, and personalisation in exoskeleton-assisted gait rehabilitation, providing a promising direction for future intelligent rehabilitation systems.
Ali Foroutannia, Masoud Mohammadian, K. Munasinghe· Italian National Conference...· 0 citations
The findings indicate that integrating reinforcement learning with Sigmoid parameter adaptation provides a systematic and effective solution for adaptive compliance regulation in mobile exoskeleton systems, enhancing adaptability, safety, and functional relevance for stroke patients undergoing lower limb rehabilitation training.
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.
Weitian He, Xinhao Zhang, Qinchen Yang et al.· IEEE Transactions on Automat...· 0 citations
Safe and intuitive human robot interaction (HRI) requires precise regulation of contact forces and torques while adapting to dynamic and uncertain human behavior. Traditional impedance and admittance control strategies rely on fixed parameters and accurate system modeling, which often limit their performance in unstructured or collaborative environments. This paper presents an AI-enabled force and torque control framework that integrates machine learning techniques with conventional control methods to enhance adaptability, compliance, and safety in physical human robot interaction. The proposed approach employs deep neural networks and reinforcement learning to learn human intent and interaction dynamics directly from multi-modal sensor data, including force torque sensors, joint encoders, and inertial measurements. By continuously adjusting control gains in real time, the system achieves stable interaction while minimizing excessive contact forces and undesired torques. Experimental evaluations conducted on a collaborative robotic platform demonstrate significant improvements over classical control schemes, including reduced interaction force peaks, smoother torque profiles, and improved task execution efficiency during cooperative manipulation tasks. The results indicate that AI-driven force and torque control can substantially improve robustness, adaptability, and user comfort in human robot collaboration, making it a promising solution for applications in rehabilitation robotics, assistive devices, and industrial cobots.
Vishal Khanna· i-manager's Journal on Augme...· 0 citations
Knee rehabilitation robots can effectively assist the recovery of patients with lower limb motor dysfunction. This paper presents a lightweight single-drive adaptive knee rehabilitation robot for supine training. The robot consists of a waist support, a thigh mechanism, and a shank-foot mechanism, and provides continuous adjustment of thigh length, shank length, and foot support angle to improve geometric matching for users with different body sizes and postures. To reduce the influence of the device’s self-weight on training, a gravity compensation method based on kinematic coupling is developed, and a sensorless interaction torque estimation method based on motor current feedback is introduced. Using only a single encoder and motor current feedback, the system achieves passive, active-assistive, and resistive behaviors within a unified control framework, without explicit mode switching. Experimental results show that, in active-assist mode, the initial response delay is about 240.0 ms, the trajectory correlation coefficient reaches 0.9960, and the maximum position error is 13.48 mm. In admittance-based interaction control mode, the estimated interaction torque agrees well with the measured torque, with a Pearson correlation coefficient of 0.954, and the average volunteer rating under six impedance levels remains above 3.7/5. These results demonstrate that the proposed robot achieves good adaptability, stable tracking, and reliable human-robot interaction, providing a practical solution for low-cost and integrated knee rehabilitation.
Mei Feng, Xiang Chen, Chao Han et al.· 2026 IEEE International Conf...· 0 citations
The results demonstrate that the proposed framework improves contact transition safety, force regulation accuracy, and task adaptability at the control-performance level, providing a feasible basis for further development of robotic physiotherapy systems.
Xueqian Zhai, Qihao Feng, Haochen Zheng et al.· Frontiers in Robotics and AI· 0 citations
