Exoskeleton Torque Control Without Gait Phase Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing exoskeleton control systems are complex and inefficient, failing to effectively minimize metabolic expenditure during versatile activities like walking and running due to their reliance on gait phase estimation and task classification, which limits their generality and efficiency.
Innovation Solution
An end-to-end, deep reinforcement learning-based control method that trains a deep neural network to provide continuous torque assistance without phase estimation or task classification, using offline simulation to transfer a hip exoskeleton system, incorporating muscle-actuated human controllers and dynamic randomization to adapt to individual variability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If gait phase estimation and task classification are used in exoskeleton control, then control precision is improved, but device complexity increases
Solution Approach 1:
The patent extracts and removes the complex gait phase estimation and task classification modules from the control system. Instead of using these intermediate processing steps, the system directly maps sensor inputs to control outputs through a simplified neural network architecture, thereby reducing device complexity while maintaining control effectiveness
Solution Approach 2:
The patent merges multiple control functions into a single end-to-end neural network. The network simultaneously performs sensing, decision-making, and control generation in one integrated architecture, eliminating the need for separate gait phase estimation and task classification systems, thus reducing overall device complexity
2Measurement precision
If gait phase estimation and task classification are used in exoskeleton control, then control precision is improved, but adaptability deteriorates
Solution Approach 1:
The patent implements a universal control system that can handle multiple gait modes (walking, running, transitioning between them) and various terrains through a single neural network architecture. The end-to-end learning approach enables the system to adapt to different activities without requiring separate control parameters or classification schemes, thereby improving generality and versatility
Solution Approach 2:
The patent employs a dynamic control approach where the neural network continuously adapts its output based on real-time sensor inputs rather than relying on pre-defined gait phases. This dynamic adaptation allows the system to seamlessly handle transitions between different gait modes and activities, enhancing versatility without needing explicit task classification
3Measurement precision
If hierarchical torque control rules with many control parameters are used, then control precision is improved, but device complexity increases
Solution Approach 1:
The patent extracts and removes the hierarchical control structure with multiple tuning parameters. Instead of using layered control rules requiring extensive parameter adjustment, the system employs a single neural network that learns optimal control policies end-to-end, significantly reducing the number of control parameters while maintaining torque control accuracy
Solution Approach 2:
The neural network in the patent performs self-learning and self-adjustment through end-to-end training, automatically optimizing control parameters during the learning process without requiring manual tuning. This self-service capability eliminates the need for complex parameter specification and adjustment, reducing device complexity while maintaining control precision
4Adaptability or versatility
If end-to-end deep reinforcement learning is used, then adaptability is improved, but use of energy increases
Solution Approach 1:
The patent implements a feedback mechanism where the neural network continuously receives sensor inputs from the exoskeleton and wearer, processes this real-time information, and adjusts torque assistance accordingly. This closed-loop feedback enables adaptive control that responds to actual physiological and mechanical conditions, optimizing energy assistance while maintaining adaptability across different activities
Data Source
AI summary
Various examples are provided related to continuous control of exoskeletons. In one example, a method includes obtaining IMU sensor signals associated with an exoskeleton attached to a limb of a subject; generating an exoskeleton control signal in response to the IMU sensor signals, the exoskeleton control signal generated by a control policy neural network trained offline from the exoskeleton using musculoskeletal human modeling and exoskeletal modeling with dynamics randomization; and controlling joint torques of the exoskeleton exerted on the subject based upon the exoskeleton control signal.