Development of a Cost-Constrained Intelligent Prosthetic Knee with Real-Time Machine Learning, Predictive Stumble Control

Lucas Jonathan Galey, University of Texas at El Paso


The field of biomechatronics is evolving quickly with advances in computer science, biology, and electrical and mechanical engineering. Coupled with increased interests in machine learning (ML) across all industry sectors, there are opportunities to leverage advanced analytics in uniquely complex problems. This study aimed to deploy real-time ML predictions in a novel microprocessor-controlled prosthetic knee (MPK) device capable of identifying and responding to stumble-events to reduce amputee fall prevalence. Innately, stumbling is a chaotic event. Current MPKs operate by detecting gait characteristics and reacting to preprogrammed states. While these systems are beneficial in significant ways, such as energy expenditure and stability, chaotic events can mislead traditional gait interpretation methods. A novel method was designed to implement an ML model capable of predicting amputee stumble occurrences using Long-Short Term Memory (LSTM) architecture on three-dimensional acceleration and velocity obtained from inertial measurement units (IMUs). This innovative approach had four main aims: (1) develop a cost-constrained prototype MPK, the GKnee; (2) collect in vivo stumble-induction data and train an ML model for the unique use-case; (3) create a control system to use the ML model to incorporate into the GKnee; and (4) combine and test components in a comprehensive stumble-induction study against a passive mechanical knee and industry standard MPKs.For the first aim, the prototype GKnee was designed to bear a 100 kg patient during heavy loading activities with an additional safety factor (three times body weight). The developed prosthetic knee does not limit flexion and dampening does not prevent normal leg swing. It was equipped with accelerometers, gyroscopes, and a servo to allow data collection and microprocessor control. The final system cost was $1,392, which was approximately $900 above the hypothesized cost.Towards the second aim, an initial patient trial was conducted with three subjects and each was induced to stumble under three different methods: bungee (n=42), obstacle (n=43), and uneven surface (n=65). With an IMU-equipped knee, the data from these trials was recorded and labeled for ML training. An LSTM ML model was then developed to classify the data and achieved an average step accuracy of 66.9%.Building on the ML model, the third aim involved developing a novel control system that switched between “walking” and “stumbling” conditions in real-time to control the GKnee with an electronically actuated hydraulic system. The control system was able to change between true states in less than 0.15 seconds and switch out of false states in less than 0.034 seconds. Additionally, the system interpreted and improved the classification rates of the ML model and increased the step accuracy of the system to 91.4%.As part of the fourth aim, a final five-subject in vivo trial was conducted that compared the GKnee to industry standard MPK systems and a mechanical passive system during stumble inductions. A total of 500 stumbles were induced across all patients, knees, and stumble induction modes. With a statistical significance of p = 0.0422, the GKnee was found to have a large effect improvement over both the MPK (Cohen’s d = 1.63) and M3 (Cohen’s d = 1.91) in stumble recovery rate. An error during data collection made it clear that this improvement resulted purely from the mechanical portion of the GKnee with no active control methods. Thus, while the ML model and control system performed according to their objectives, they ultimately had no effect on the improved recovery rate during testing.Overall, this research contributes an in-depth, transfemoral-amputee stumble dataset with an effective ML model network while also presenting a cost-constrained prosthetic knee prototype that effectively reduces amputee fall incidence rate during treadmill induced stumbles compared to industry standard devices.

Subject Area

Biomedical engineering|Computer science|Health sciences|Biomechanics|Artificial intelligence

Recommended Citation

Galey, Lucas Jonathan, "Development of a Cost-Constrained Intelligent Prosthetic Knee with Real-Time Machine Learning, Predictive Stumble Control" (2023). ETD Collection for University of Texas, El Paso. AAI30521679.