In dense forests or gravel roads, we humans use an array of senses to move freely. We hear branches rustling, feel the texture of the ground and sense our own balance. Until now, robots only saw, but that is changing.
Human perception to robots with sensory intelligence
Researchers at Duke University have developed an artificial intelligence system for artificial intelligence system that allows quadruped robots to sense their environment with a combination of vision, touch, and vibration. Called WildFusion, this framework fuses sensory data to improve autonomous navigation in complex terrain.
WildFusion integrates an RGB camera, LiDAR sensorssensors, contact microphones, inertial and tactile sensors. While walking, the robot picks up ground vibrations, measures the pressure of its legs and evaluates its inclination with each step. This wealth of data makes it possible to anticipate whether a surface is stable or slippery.
Real-time data processing
Each type of information is processed by specialized encoders that feed a deep learning model. Instead of working with isolated points, the system builds a continuous representation of the terrain, allowing for more natural movement decisions.
The system was evaluated at Eno River State Park in North Carolina. The robot demonstrated a remarkable ability to adapt to different types of terrain: from dry leaves to mud to gravel. This test demonstrated its ability to predict trafficability with high accuracy.
In addition to assessing the stability of surfaces, the system was able to recognize acoustic terrain patterns, allowing the robot to anticipate sudden changes in soil texture or consistency. This level of perception improves route planning and reduces errors in dynamic environments.
Possibilities of use in critical contexts
The architecture of WildFusion is modular, making it easy to incorporate thermal or humidity sensors in future versions. This makes it a viable option for tasks in disaster areas, remote inspections or exploration in natural environments where visual perception alone is insufficient.
With this technology, robotics takes a step towards systems that not only observe, but also sense and understand the world.
Follow us on social networks and don’t miss any of our publications!
LinkedIn Facebook Instagram X (Twitter) TikTok
Source and photo: Duke University
