Integrated Barometric Pressure Sensors on Legged Robots for Enhanced Tactile Exploration of Edges

Thijs Van Hauwermeiren, Anatolii Sianov, Annelies Coene, Guillaume CrevecoeurIEEE RA-L 2024 · DOI

Introduction

As legged robots transition from controlled laboratory settings to complex real-world environments like construction sites and industrial facilities, the need for precise terrain perception becomes critical. While vision-based systems such as LiDAR and depth cameras are standard for global mapping, they frequently encounter "blind spots" in the immediate vicinity of the robot's feet. Furthermore, these sensors can be compromised by challenging lighting conditions or physical occlusions.

To address these limitations, researchers have developed a rugged, low-cost tactile foot designed for quadruped robots. This "smart foot" enables robots to "feel" their environment, specifically identifying edges and terrain features with high precision. This tactile feedback is essential for safe navigation, ensuring the robot does not commit its weight to unstable or non-existent footholds.

Conceptual overview of tactile exploration and edge detection.

Sensor Design and Hardware

The core of the innovation lies in a sensor assembly that is both sensitive and exceptionally durable. The hardware consists of eight MEMS barometric pressure sensors (MS5840-02BA) arranged in a circular pattern on a custom PCB. This electronic core is housed within a 3D-printed frame and protected by a soft, deformable rubber half-sphere (Shore 15A).

Exploded view of the tactile foot assembly: (1) base, (2) PCB with sensors, (3) spacer, (4) retaining ring, and (5) deformable rubber dome.

When the foot makes contact with a surface, the rubber dome deforms, causing internal air pressure changes. These changes are captured by the sensor array. This design leverages mechanical compliance—the "sole" specifically stretches and deforms around edges—to capture a unique pressure "signature" of the terrain. Using air pressure instead of traditional strain gauges makes the sensor robust enough to survive repeated 100N impacts, typical for a 12kg quadruped like the Unitree Go1.

Contact Classification and Edge Regression

The system processes the 8-channel pressure data through a hybrid framework to classify the contact type and estimate edge orientation.

Contact Classification

To determine the nature of the contact, the system utilizes two primary statistical features:

  • Mean Pressure ($\mu$): Used to detect the occurrence of contact.
  • Coefficient of Variation ($CV$): Calculated as the standard deviation of the sensors divided by the mean. This scale-invariant metric is crucial because it allows the system to identify edges regardless of the total force applied (which varies with robot gait or payload). A low $CV$ indicates a flat surface with even pressure distribution, while a high $CV$ signifies an edge.

Edge Regression

Once an edge is detected, the raw pressure signals are fed into a fully connected Neural Network (NN). This model is trained to predict the orientation ($\alpha$) of the edge relative to the foot. The training data was generated using a UR5 cobot arm to perform hundreds of controlled "pokes" against various edge orientations and distances, creating a comprehensive database of 256 unique contact points.

Experimental Results

In controlled laboratory evaluations, the sensor demonstrated remarkable accuracy, predicting edge orientation with a Mean Absolute Error (MAE) of just 2°. The system's robustness was further validated under several "real-world" conditions:

  • Payload Stability: Accuracy remained consistent even with an additional 3.75kg payload (30% of the robot's mass).
  • Steep Inclines: The sensor successfully detected edges on slopes up to 75°.
  • Dynamic Disturbances: Classification remained reliable even when the robot was subjected to 100N external forces.

When compared to standard scalar force sensors, which only measure total vertical force, this tactile foot significantly improved safety by detecting edges before the robot attempted to step on them.

Conclusion and Future Work

The integration of barometric pressure sensors into a deformable foot provides a robust and high-resolution solution for tactile terrain perception. While currently optimized for straight edges, future iterations will aim to detect more complex geometries like corners. Additionally, the researchers plan to implement "proprioceptive-tactile fusion," combining foot-level data with joint sensors and vision systems to create a more holistic and resilient navigation stack for all-terrain exploration.

Reference: Van Hauwermeiren, T., et al. "Integrated Barometric Pressure Sensors on Legged Robots for Enhanced Tactile Exploration of Edges." IEEE Robotics and Automation Letters, 2024.

Summary updated

BibTeX
@article{hauwermeiren2024integrated,
  author = {Thijs Van Hauwermeiren and Anatolii Sianov and Annelies Coene and Guillaume Crevecoeur},
  title = {{Integrated Barometric Pressure Sensors on Legged Robots for Enhanced Tactile Exploration of Edges}},
  journal = {IEEE RA-L},
  year = {2024},
  doi = {10.1109/lra.2024.3405378},
}
Figure