Traversal by Touch: Tactile-Based Robotic Traversal with Artificial Skin in Complex Environments

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Fusion

Background:

• Robotic navigation often relies on visual sensors, which are susceptible to varying lighting conditions.
• Developing robust navigation systems for environments with poor visibility (darkness, smoke) is a significant challenge.
• Tactile sensing offers a potential alternative or complementary modality for robot locomotion.

Purpose of the Study:

• To evaluate the performance of tactile-first robotic traversal algorithms on the DHS figure-8 mobility test.
• To compare the robustness of different tactile policies against vision-based and tactile histogram baselines under diverse lighting conditions.
• To assess the feasibility of lighting-independent navigation using tactile sensing.

Main Methods:

• A two-way repeated-measures design was used, testing multiple algorithms (M1, M2, M3, CB-V, T-VFH, T-D* Lite) across Indoor, Outdoor, and Dark lighting conditions.
• The custom-built Eleven robot, a quadruped with joint-mounted tactile sensors (FSR, CNT yarn), was utilized for testing.
• Wireless control and sensing were implemented using ESP32-S3, Arduino Nano 33 BLE, Raspberry Pi 400, and a mini VESC controller.

Main Results:

• The tactile stack achieved approximately 21 ms policy latency (p50) and mid-80% success rates across all lighting conditions, including complete darkness.
• The memory-augmented tactile policy (M3) demonstrated consistent robustness compared to the camera baseline (CB-V), with performance degradation of only 3-4% indoors and 13-16% outdoors/dark.
• No speed equivalence was confirmed between M3 and CB-V, indicating a trade-off between speed and robustness.

Conclusions:

• Tactile-first traversal, particularly with a memory-augmented policy, provides lighting-independent navigation capabilities.
• This approach is essential for robots operating in environments with challenging illumination or texture conditions.
• The study highlights the potential of tactile sensing for robust and reliable robotic navigation, trading modest speed for enhanced sensing independence.