Bio-inspired risk-aware adaptive navigation for quadruped robots in search and rescue scenarios

Search and Rescue (SAR) operations demand robotic systems capable of navigating hazardous and unstructured environments where human intervention is limited. This work introduces a bio-inspired navigation framework that enables quadruped robots to adapt their trajectories in real time based on perceived environmental risk. The approach integrates visual and depth perception with convolutional neural networks (CNN) to identify terrain features and estimate risk levels, generating dynamic maps that guide adaptive waypoint planning. A key aspect of this research is the extensive simulation campaign carried out in highly realistic post-disaster environments, including conditions with fire, smoke, debris, uneven surfaces, and degraded lighting. These scenarios provide a controlled yet challenging platform for validating adaptive behaviors before deployment in real missions. The proposed method is further evaluated in physical trials, confirming its ability to enhance safety, robustness, and efficiency compared to static navigation strategies. This study demonstrates the potential of combining bio-inspired principles with intelligent perception to enhance autonomous mobility in critical missions. The implemented algorithms and video of the testing phase are available in the Appendix.