TL;DR
This paper presents the first model-dependent prosthesis knee controller that uses real-time force sensing from both the human-prosthesis interface and the ground. This approach achieves stable walking on various terrains by directly incorporating measured interaction forces into the control loop, leading to significantly improved tracking performance compared to controllers that estimate or ignore these forces.
What Problem Does This Paper Address?
Lower-limb prosthesis users experience a higher rate of falls compared to non-amputees, partly due to the inherent instability of passive prostheses. While powered prostheses offer improved stability, current control methods often rely on heuristic tuning, lack formal stability guarantees, and fail to fully leverage the natural dynamics and force interactions between the user, the prosthesis, and the environment. Existing model-based approaches have struggled with the unknown human dynamics and the computational cost of real-time force sensing and inversion. This paper aims to bridge this gap by developing a robust, model-dependent controller that utilizes real-time force sensing to achieve stable and adaptable locomotion across diverse terrains.
Methods
The core of this work is the development and implementation of a model-dependent prosthesis knee controller that integrates real-time force sensing. The controller is built upon the framework of Rapidly Exponentially Stabilizing Control Lyapunov Functions (RES-CLFs), specifically an ID-CLF-QP (Inverse Dynamics Control Lyapunov Function - Quadratic Program) formulation. This approach allows for formal stability guarantees by leveraging the system's dynamics.
Key methodological innovations include:
- Real-time Force Sensing Integration: The controller directly uses measurements from:
- A load cell at the human-prosthesis interface to measure socket forces ($F_f$).
- An insole pressure sensor to estimate vertical ground reaction force ($F_{g,z}$) and pitch ground reaction moment ($M_{g,y}$).
- An IMU (Inertial Measurement Unit) for orientation and angular velocity in non-stance phases.
- Inverse Dynamics Framework: To avoid computationally expensive matrix inversions, the controller is formulated within an inverse dynamics framework, realized as a Quadratic Program (QP). This allows for the inclusion of measured forces and moments directly into the dynamics.
- Separable Subsystem Control: The human-prosthesis system is modeled as a separable subsystem, where the prosthesis dynamics are controlled using local information, relying on the assumption that the human provides stable locomotion.
- Hybrid Zero Dynamics for Gait Generation: Trajectories are generated using hybrid zero dynamics optimization, ensuring invariance through impact events and matching human walking data.
Datasets
Not discussed in the paper. The paper utilizes data from human subjects performing walking tasks on various terrains, but these are presented as experimental results rather than pre-defined datasets for training. Motion capture data from [29] was used to generate desired trajectories.
Key Insights and Ideas
- Force Sensing is Crucial for Model-Dependent Control: Directly measuring interaction forces (socket and ground) significantly improves the accuracy of the system model, enabling the controller to adapt to real-world conditions and achieve better tracking performance.
- Model-Dependent Control Offers Adaptability: By accounting for the system's dynamics and external forces, the model-dependent controller demonstrates better performance across different terrains and subjects without requiring extensive re-tuning, suggesting greater transferability.
- Real-time Force Sensing Enables Stability Guarantees: Integrating measured forces into the control loop strengthens the theoretical basis for stability guarantees, making the system more reliable for users.
- QP Formulation Simplifies Implementation: Realizing the RES-CLF controller via a QP formulation overcomes computational challenges and makes it practical for hardware implementation.
Key Results
- Improved Tracking Performance: The force-sensing ID-CLF-QP controller achieved significantly lower Root Mean Square Error (RMSE) in stance phase tracking compared to controllers without force sensing, force estimation, or a traditional PD controller.
- For example, on a rubber floor, the RMSE for the force-sensing controller was 0.0409 rad, while the no-sensor controller was 0.0797 rad (Table I).
- Across four terrains and two subjects, the force-sensing controller consistently showed the lowest stance RMSE (Table II).
- Robustness Across Terrains: The controller demonstrated stable walking and improved tracking on four different terrains: rubber floor, outdoor track, grass, and sidewalk.
- Transferability: The controller showed better performance for two subjects without re-tuning, indicating its potential for broader applicability across users.
- Stable Human-Prosthesis Walking: Experimental results confirmed stable human-prosthesis walking with the implemented controller.
Limitations and Future Work
- Limited to Knee Control: The current implementation focuses on controlling only the prosthesis knee joint. The ankle is controlled by a simpler PD controller.
- Single Terrain Type for Subject 2: Subject 2 was tested on only one terrain type, limiting the scope of multi-terrain evaluation for that subject.
- Potential for Human-Prosthesis Mismatch: While the human is assumed to stabilize itself, the paper acknowledges that the human's response to the prosthesis's dynamic behavior could be further investigated.
- Future Work:
- Full-Body Control: Extend the force-sensing ID-CLF-QP controller to both the knee and ankle for more complex gait patterns, including heel-toe roll.
- Amputee Subject Testing: Evaluate the controller with amputee subjects to assess its real-world applicability.
- Multi-Contact Gait: Apply the controller to more complex multi-contact walking scenarios.
- Investigate Human Dynamics: Further explore the human's role in maintaining stability and how it interacts with the prosthesis.
Summary updated