TSLAM: Tethered simultaneous localization and mapping for mobile robots

Abstract

Tethered mobile robots are useful for exploration in steep, rugged, and dangerous terrain. A tether can provide a robot with robust communications, power, and mechanical support, but also constrains motion. In cluttered environments, the tether will wrap around a number of intermediate ‘anchor points’, complicating navigation. We show that by measuring the length of tether deployed and the bearing to the most recent anchor point, we can formulate a tethered simultaneous localization and mapping (TSLAM) problem that allows us to estimate the pose of the robot and the positions of the anchor points, using only low-cost, nonvisual sensors. This information is used by the robot to safely return along an outgoing trajectory while avoiding tether entanglement. We are motivated by TSLAM as a building block to aid conventional, camera, and laser-based approaches to simultaneous localization and mapping (SLAM), which tend to fail in dark and or dusty environments. Unlike conventional range-bearing SLAM, the TSLAM problem must account for the fact that the tether-length measurements are a function of the robot’s pose and all the intermediate anchor-point positions. While this fact has implications on the sparsity that can be exploited in our method, we show that a solution to the TSLAM problem can still be found and formulate two approaches: (i) an online particle filter based on FastSLAM and (ii) an efficient, offline batch solution. We demonstrate that either method outperforms odometry alone, both in simulation and in experiments using our TReX (Tethered Robotic eXplorer) mobile robot operating in flat-indoor and steep-outdoor environments. For the indoor experiment, we compare each method using the same dataset with ground truth, showing that batch TSLAM outperforms particle-filter TSLAM in localization and mapping accuracy, owing to superior anchor-point detection, data association, and outlier rejection.

Developing and deploying a tethered robot to map extremely steep terrain

“…Next, the robot can either repeat back along the same path to unwind from added anchors McGarey, Polzin, and Barfoot () or continue to explore safe paths that expand the frontier of the estimated map. It would also be beneficial to integrate autonomous path following with anchor detection, where McGarey, MacTavish, Pomerleau, and Barfoot () details a novel SLAM technique for anchors, which could potentially allow for extending Visual Teach & Repeat (VT&R) to work within safe, boundary regions defined by detected anchors. Most importantly, the operator can monitor mapping progress and relax while the robot autonomously explores all the accessible terrain from a given anchor location.…”

Exactly sparse Gaussian variational inference with application to derivative-free batch nonlinear state estimation

“…By using only bearing measurements, this proved to be a challenging dataset. We initialized our landmark locations using the bearing-only random sample consensus (RANSAC) (Fischler and Bolles 1981) strategy described by McGarey et al (2017). We attempted to initialize our robot states using only the wheel odometry information, but this proved too difficult for methods making use of the full Hessian (i.e., Newton’s method).…”

Navigation on the Line: Traversability Analysis and Path Planning for Extreme-Terrain Rappelling Rovers

“…We fielded several aspects on the Axel rover at a Mars analogue site, which demonstrate first steps towards developing an autonomous tethered platform. Future work would mature these algorithms to address uncertainties in sensing and control that include: i) field testing of tetherconstraints on physical platforms, iii) developing a tetheraware controller, iii) developing a tether-and anchor-aware pose estimation that builds on prior work by [29], and iv) leveraging global-map knowledge from orbital data into the onboard navigation system.…”