Exoskeleton Gait-Phase Control With Real-Time CNN Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current exoskeleton technologies face challenges in modulating assistance across various ambulation modes and intensities without extensive context-specific tuning, limiting their effectiveness in real-world applications and maintaining a gap between lab-based and real-world benefits.
Innovation Solution
A unified exoskeleton control framework using a convolutional neural network (CNN) and temporal convolutional network (TCN) for autonomous adaptation based on instantaneous user joint moment estimates, which reduces metabolic cost and lower-limb positive work across different ambulation conditions without subject-specific calibration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If powered exoskeletons use motors or actuators to provide mechanical assistance, then mobility assistance for individuals with mobility impairments is improved, but power consumption increases limiting range and endurance
Solution Approach 1:
The exoskeleton system dynamically adapts its assistance level based on real-time detection of user intent and contextual information. The control system modulates motor output dynamically, providing high assistance when needed and reducing assistance when not required, thereby improving mobility support while managing power consumption throughout the operational period
Solution Approach 2:
The system continuously monitors sensor data from the user's movement patterns, muscle activity, and exoskeleton performance to provide real-time feedback to the control algorithm. This feedback loop enables the system to adjust assistance levels optimally, ensuring effective mobility support while avoiding unnecessary power consumption
2Reliability
If powered exoskeletons are optimized for particular modes of ambulation, then assistance effectiveness is improved, but adaptability to changing or uneven terrain deteriorates
Solution Approach 1:
The exoskeleton employs dynamic control that continuously adapts to changing terrain conditions and user needs. The system detects transitions between different ambulation modes and adjusts its assistance strategy in real-time, maintaining reliability across varied terrains without requiring pre-optimization for specific conditions
Solution Approach 2:
The control system is designed with universal adaptability to handle multiple ambulation modes and terrain types through a single unified framework. By using contextual information and machine learning algorithms, the system can effectively assist across diverse conditions including level ground, inclines, and uneven terrain without requiring separate optimization for each scenario
3Use of energy by moving object
If exoskeletons provide substantial mechanical work assistance, then human energetics are improved, but device complexity increases limiting real-world deployment
Solution Approach 1:
The control system is segmented into modular functional components including sensor processing modules, intent detection algorithms, and assistance generation modules. This segmentation allows the complex control task to be divided into manageable parts that can be independently optimized and maintained, reducing the barrier to real-world deployment while preserving the ability to substantially assist human energetics
Solution Approach 2:
The exoskeleton system uses unsupervised machine learning to automatically learn and adapt to individual user characteristics and movement patterns without requiring extensive manual calibration or tuning. This self-service capability reduces the complexity of deployment and customization, enabling the system to provide substantial mechanical work assistance while simplifying the control architecture
Data Source
AI summary
A device may include a high-level control layer comprising a convolutional neural network (CNN) configured to receive exoskeleton sensor data from one or more sensors on an exoskeleton and generate a user state estimate. The device may include a mid-level control layer configured to receive the user state estimate and generate a torque command for an actuator based on the user state estimate. The user state estimate may be an estimated gait phase, where the mid-level control layer generates the torque command as a function of the estimated gait phase based on an assistance profile. The high-level control layer may include a backward labeler and a real-time adaptation trainer. The backward labeler relabels ground truth gait phase from the exoskeleton sensor data using a local peak detection. The real-time adaptation trainer trains the CNN in a single epoch of backpropagation with the ground truth gait phase.