Exoskeleton Torque Control via Hybrid Feedback and Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current exoskeleton torque control methods face challenges in accurately tracking desired torques during human-robot interactions, particularly due to complex and changing dynamics, with limited comparative studies across different controllers and platforms, leading to difficulties in selecting and tuning effective control strategies for lower-limb exoskeletons.
Innovation Solution
The development of a system that compares the torque-tracking performance of prominent torque controllers, including classical feedback, model-based, adaptive, and iterative learning control methods, on a single exoskeleton platform, using high-level controllers based on time, joint angle, neuromuscular models, and electromyography measurements, with a combination of proportional control, damping injection, and iterative learning resulting in lower root-mean-squared errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If classical feedback control is used for torque tracking, then the control structure is simple, but the torque tracking precision is insufficient under uncertain and changing human dynamics
Solution Approach 1:
The patent implements a hierarchical feedback control structure where an outer loop uses proportional control to generate torque commands based on position error, and an inner loop uses derivative control with damping injection to track the desired torque. This nested feedback approach improves torque tracking precision while maintaining reasonable structural simplicity.
Solution Approach 2:
The patent applies iterative learning control to pre-compute feedforward torque commands based on desired trajectory information before execution. This preliminary action anticipates required torques, allowing the feedback loops to focus on correcting deviations, thereby improving tracking precision without proportionally increasing real-time control complexity.
2Measurement precision
If model-based control is used to improve torque tracking, then the tracking precision improves, but the difficulty of detecting and measuring system parameters increases
Solution Approach 1:
The patent transforms the complex model-based control problem into parameter tuning of a simplified proportional-derivative structure. Instead of requiring full system identification and complex model inversion, the approach focuses on tuning proportional gain Kp and derivative gain Kd to achieve optimal torque tracking, significantly reducing measurement and detection requirements.
Solution Approach 2:
The patent replaces complex, computationally intensive model-based controllers with a simpler proportional-derivative structure that requires fewer and less precise sensors. This substitution achieves comparable or superior performance with reduced system parameter detection requirements.
3Adaptability or versatility
If adaptive control is used to handle changing dynamics, then the adaptability improves, but the device complexity increases
Solution Approach 1:
The patent applies iterative learning control that periodically updates feedforward torque commands based on performance from previous gait cycles. This periodic adaptation exploits the cyclic nature of walking to improve performance over time without requiring continuous complex adaptive algorithms, maintaining relatively simple controller structure while achieving good adaptability.
Solution Approach 2:
The controller uses its own past performance data to automatically adjust future control actions through iterative learning. The system self-tunes by comparing actual torque with desired torque from previous cycles and using this information to improve subsequent cycles, reducing the need for external tuning and complex adaptive mechanisms.
4Measurement precision
If iterative learning control is used for cyclic walking tasks, then the torque tracking precision improves, but the loss of time for learning increases
Solution Approach 1:
The patent implements a hybrid controller that combines iterative learning (partial action) with real-time proportional-derivative feedback. The iterative learning provides improving feedforward commands over cycles, while the PD feedback ensures acceptable torque tracking from the first cycle without waiting for learning convergence, effectively distributing the precision achievement across time and reducing perceived learning time.
Solution Approach 2:
The proportional-derivative feedback control operates continuously from the first cycle, providing useful torque tracking improvement immediately, while iterative learning operates in parallel to progressively enhance performance. This continuous useful action ensures that time is not lost waiting for learning to converge, as feedback control provides immediate benefits.
Data Source
AI summary
This document describes systems and methods for controlling an exoskeleton. The system receives a measurement of a first torque applied to a rotational joint coupling a first component to a second component, the first torque being applied by a motor via a cable. The system determines, based on the measurement of the first torque, a first portion of a second torque to apply to the rotational joint. The system determines, based on the measurement of the first torque, a second portion of the second torque to apply to the rotational joint. The system determines a value of the second torque to apply to the rotational joint based on the first portion and the second portion. The system controls the motor for applying the second torque to the rotational joint via the cable.