Autonomous Off-road Navigation over Extreme Terrains with Perceptually-challenging Conditions

We propose a framework for resilient autonomous navigation in perceptually\nchallenging unknown environments with mobility-stressing elements such as\nuneven surfaces with rocks and boulders, steep slopes, negative obstacles like\ncliffs and holes, and narrow passages. Environments are GPS-denied and\nperceptually-degraded with variable lighting from dark to lit and obscurants\n(dust, fog, smoke). Lack of prior maps and degraded communication eliminates\nthe possibility of prior or off-board computation or operator intervention.\nThis necessitates real-time on-board computation using noisy sensor data. To\naddress these challenges, we propose a resilient architecture that exploits\nredundancy and heterogeneity in sensing modalities. Further resilience is\nachieved by triggering recovery behaviors upon failure. We propose a fast\nsettling algorithm to generate robust multi-fidelity traversability estimates\nin real-time. The proposed approach was deployed on multiple physical systems\nincluding skid-steer and tracked robots, a high-speed RC car and legged robots,\nas a part of Team CoSTAR's effort to the DARPA Subterranean Challenge, where\nthe team won 2nd and 1st place in the Tunnel and Urban Circuits, respectively.\n