TL;DR
This paper introduces a novel tactile admittance control system for quadruped robots that directly measures ground reaction forces (GRFs) using spherical foot sensors. This allows for more accurate and immediate feedback, leading to significantly improved balance and stability during disturbances and locomotion compared to traditional methods.
What Problem Does This Paper Address?
Quadruped robots rely heavily on ground reaction forces (GRFs) for balance and locomotion. However, most existing control systems estimate these forces indirectly using joint torques and kinematic models. This indirect estimation is prone to errors due to assumptions and modeling inaccuracies, making robots less robust to unexpected disturbances. The paper addresses this by proposing a system that directly senses contact forces at the feet using tactile sensors, enabling more precise control and richer environmental interaction.
Methods
The core of the proposed system involves three main contributions:
- Fast 3D Force Reconstruction: A computationally efficient method to reconstruct 3D forces from pressure readings of spherical tactile sensors. This method uses a linear model calibrated with force-torque measurements, allowing for real-time application.
- Tactile Admittance Controller: This controller integrates the measured contact forces into the robot's leg control loop. It adjusts leg motion to achieve desired GRFs and compliance, effectively acting as a virtual mass-spring-damper system that reacts to external forces.
- Experimental Validation: The system was tested on a Unitree A1 quadruped robot equipped with custom spherical tactile feet. Experiments included static balance on a beam and a dynamic jumping-and-landing scenario.
Datasets
Not discussed in the paper. The paper focuses on experimental validation using a custom-built setup and a quadruped robot.
Key Insights and Ideas
- Direct Force Sensing is Key: Directly measuring GRFs with tactile sensors offers a significant advantage over indirect estimation methods, leading to more accurate and responsive control.
- Spherical Tactile Sensors are Versatile: The proposed force reconstruction method works well with spherical tactile sensors, which are robust and can provide wide-area sensing coverage.
- Admittance Control for Robustness: Integrating tactile feedback into an admittance control framework allows the robot to dynamically adapt its leg movements to external forces, enhancing stability and compliance.
- Simplified Control Design: By directly measuring forces, the need for complex dynamic models, actuator parameter identification, and contact normal estimation is reduced, making the system more portable and easier to implement.
Key Results
- Static Balance: In a static balance experiment on a beam, the tactile admittance control reduced the external disturbance force by 23% and the peak ground reaction forces on the overloaded leg by 55%.
- Dynamic Balance: During a dynamic jumping-and-landing scenario, the robot with tactile admittance control stabilized effectively after impact, whereas the baseline PD controller failed, leading to a trip and fall.
- Force Reduction: The peak force on an overloaded leg was reduced from 90 N to 40 N (a 55% reduction) with admittance control.
- Improved Load Distribution: The admittance control led to a more balanced redistribution of load across all feet during external disturbances.
Limitations and Future Work
- Limited Compliance Direction: The current implementation primarily focused on compliance in the vertical (z) direction.
- Fixed Control Gains: The experiments used fixed admittance control parameters.
- Mechanical Durability: The long-term mechanical durability of the soft sensor feet needs further evaluation.
Future work suggested includes:
- Extending compliance to multiple axes (x, y, and z).
- Enabling online parameter tuning for adaptive control.
- Investigating multi-directional adaptive admittance control.
- Fusing tactile data with other sensory modalities like vision or LiDAR.
- Prolonged testing under varied terrain and environmental conditions.
Summary updated