Motion- and Uncertainty-aware Path Planning for Micro Aerial Vehicles

Markus W. Achtelik, Simon Lynen, Stephan Weiss, Margarita Chli, Roland Siegwart
2014 Journal of Field Robotics  
Localization and state estimation are reaching a certain maturity in mobile robotics, often providing both a precise robot pose estimate at a point in time and the corresponding uncertainty. In the bid to increase the robots' autonomy, the community now turns to more advanced tasks, such as navigation and path planning. For a realistic path to be computed, neither the uncertainty of the robot's perception nor the vehicle's dynamics can be ignored. In this work, we propose to specifically
more » ... the information on uncertainty, while also accounting for the physical laws governing the motion of the vehicle. Making use of rapidly exploring random belief trees, here we evaluate offline multiple path hypotheses in a known map to select a path exhibiting the motion required to estimate the robot's state accurately and, inherently, to avoid motion in modes, where otherwise observable states are not excited. We demonstrate the proposed approach on a micro aerial vehicle performing visual-inertial navigation. Such a system is known to require sufficient excitation to reach full observability. As a result, the proposed methodology plans safe avoidance not only of obstacles, but also areas where localization might fail during real flights compensating for the limitations of the localization methodology available. We show that our planner actively improves the precision of the state estimation by selecting paths that minimize the uncertainty in the estimated states. Furthermore, our experiments illustrate by comparison that a naive planner would fail to reach the goal within bounded uncertainty in most cases. C 2014 Wiley Periodicals, Inc. * The research leading to these results has received funding from the European Community's Seventh Framework Programme (FP7) under grant agreements n.266470 (myCopter) and n.600958 (SHERPA). ever, we are still missing essential functionality before a robot can carry out real missions autonomously from start to finish. With current research pushing for added autonomy, realistic path planning and obstacle avoidance are certainly at the top of the priorities list. With the seminal works of Kavraki,Švestka, Latombe, and Overmars (1996) and Kuffner and LaValle (2000) paving the way for random-sampling-based path planners, the question arising is how to employ such methods within the navigation pipeline running onboard the mobile robot. While path planning has long been studied on ground vehicles, approaches dealing with MAV navigation in three dimensions are rather sparse. Perhaps this can be attributed to the fact that it is only very recently that sufficiently robust systems, able to perform simple maneuvers and estimate their motion, have started appearing (Fraundorfer et al.,
doi:10.1002/rob.21522 fatcat:z7uzvnekxba43lnhzrcz4vegmu