Robot terrain recognition improves with attention and new surface classes
Attention-Based Surface Representation Learning for Robot State Prediction and Open-Ended Surface Classification
RoboticsArtificial IntelligenceMachine Learning
Summary
For robots that move outside, knowing what kind of ground they are on helps them work better. Most methods only recognize a fixed set of ground types known beforehand. The authors created a way for robots to learn new types of ground surfaces as they encounter them. Their method uses a technique called attention and predicts the robot’s future movements to better classify surfaces. They tested this on two datasets and got very high accuracy.
What this means in practice
- •For outdoor robotics engineers: Improve robots’ ability to identify unknown terrain types encountered during outdoor navigation by expanding the set of recognized surfaces dynamically.
- •For robotics platform developers: Enhance robot perception modules to predict robot states and classify terrain more accurately for better navigation and control.
Authors
Oleg Kushnarev, Alexander Belyaev
Abstract
For ground robots operating in outdoor environments, understanding the properties of the underlying terrain is essential for ensuring reliable operation. In most perception-based studies, this problem is formulated as categorical classification with a fixed number of classes defined during training. We propose an approach that enables new surface classes to be added as trainable vectors, which can subsequently be used to address higher-level tasks. By employing a learning paradigm based on predicting the robot's next state in time and using attention blocks, we improved classification accuracy to 98.56% on the Belyaev-Kushnarev dataset and 94.8% on BorealTC.