Robot learns to predict safe paths while remembering old terrains
Continual Learning for Traversability Prediction with Uncertainty-Aware Adaptation
RoboticsArtificial IntelligenceMachine Learning
Summary
Predicting where a robot can safely travel is hard, especially on new or rough ground. The authors created a method for robots to keep learning about new surfaces without forgetting what they already know. Their approach uses a way to remember past experiences without saving all the old data and takes into account uncertainty when adapting. They tested this on a real robot and showed it can handle different terrains and avoid forgetting earlier lessons.
What this means in practice
- •For autonomous vehicle developers: Improve robots' ability to adapt to new terrains in real time without losing previously learned knowledge for safer navigation.
- •For agricultural machinery engineers: Enhance farm robots to adjust to diverse soil conditions over time while retaining operation knowledge for different environments.
Authors
Hojin Lee, Yunho Lee, Daniel A Duecker, Cheolhyeon Kwon
Abstract
Traversability prediction is a critical component of autonomous navigation in unstructured environments, where complex and uncertain robot-terrain interactions pose significant challenges such as traction loss and dynamic instability. Despite recent progress in learning-based traversability prediction, these methods often fail to adapt to novel terrains. Even when adaptation is achieved, retaining experience from previously trained environments remains a challenge, a problem known as catastrophic forgetting. To address this challenge, we propose a continual learning framework for traversability prediction that incrementally adapts to new terrains using a generative experience recall model. A key virtue of the proposed framework is two folds: i) retain prior experience without storing past data; and ii) incorporate the uncertainty of the generated samples from the recall model, enabling uncertainty-aware adaptation. Real-world experiments with a skid-steering robot validate the effectiveness of the proposed framework, demonstrating its ability to adapt across a series of diverse environments while mitigating catastrophic forgetting.