Risk-aware Integrated Task and Motion Planning for Versatile Snake Robots under Localization Failures

Abstract—Snake robots enable mobility through extremeterrains and confined environments in terrestrial and spaceapplications. However, robust perception and localization forsnake robots remain an open challenge due to the proximityof the sensor payload to the ground coupled with limitedfield-of-view. To address this issue, we propose Blind-motionwith Intermittently Scheduled Scans (BLISS) which combinesproprioception-only mobility with intermittent scans to beresilient against both localization failures and collision risks.BLISS is formulated as an integrated Task and Motion Planning(TAMP) problem that leads to a Chance Constrained HybridPartially Observable Markov Decision Process (CC-HPOMDP),known to be computationally intractable due to the curse ofhistory. Our novelty lies in reformulating CC-HPOMDP asa tractable, convex Mixed Integer Linear Program (MILP).This allows us to solve BLISS-TAMP significantly faster andderive optimal task and motion plans in a coupled manner.Simulation and hardware experiments on the EELS snake robotdemonstrate that our method shows up to 2x computationalimprovement over conventional and state-of-the-art planners.