Place Revisiting for Planetary Rovers: An Enabling Technology and Field Testing of Three Mission Concepts

Planetary rovers to date have been operated mainly in a serial mode; they are driven from one place to the next, away from the lander, and seldom return to previously visited places. We have been developing a visual navigation technique, called network of reusable paths (NRP), that can be thought of as a low-computational-cost version of simultaneous localization and mapping coupled to a path-tracking controller. The result is that a rover can be returned accurately to any place it has previously visited using only visual feedback; this enables science to be gathered from multiple sites in parallel. We will describe how NRP works, and present field test results of two mission concepts where we have made use of this technology: a lunar-sample-return scenario and a Marsmethane-hunting scenario. I. EXTENDED ABSTRACT At the time of writing, NASA’s Spirit, Opportunity, and Curiosity rovers have driven a combined 44+ kilometres on the surface of Mars, visiting many sites of scientific interest along the way. Their exploration strategies have been serial in the sense that scientific objectives are completed at one site before departing for the next. The coming decades will see more advanced missions to both Mars and the Moon. For example, the return of samples is highly desirable from both targets and could benefit from the ability to investigate multiple candidate sampling sites in parallel, in order to methodically downselect the specimens to return. To enable such a parallel investigation, rovers need the ability to accurately revisit places. We have been developing one technology that could enable place revisiting that we call network of reusable paths (NRP). NRP leverages the stereo visual odometry (VO) pipeline already standard on current rovers [1], requiring about double the computational resources of VO to enable place revisiting [2]. In essence, the localization/mapping part of NRP runs a stereo VO pipeline with two key differences. First, when establishing new paths, the visual landmarks used to compute motion (triangulated SURF features in our case) are stored relative to the robot’s path, creating a simple map. Second, when repeating a path we match the current image not only to the previous image, but also to the path-based map (see Figure 2). NRP also has a planning component based on a rapidly-expanding random tree (RRT), which allows the robot to use the existing paths it has established, but also to 1T. D. Barfoot, B.E. Stenning, J.D. Gammell, C.H. Tong, C. McManus (now at Oxford), and L.-P. Berczi are/were with the University of Toronto Institute for Aerospace Studies, 4925 Dufferin St., Toronto, Ontario, Canada, M3H 5T6, [email protected]. 2G.R. Osinski is with the Unversity of Western Ontario, Canada. 3M. Daly is with York Unversity, Canada. 4C. Dickinson is with MDA Space Missions, Canada. Fig. 1. Rover adding to a network of reusable paths on the Canadian Space Agency’s Mars Emulation Terrain; once created, it can return accurately to any previously visited place using only visual feedback. branch off into new terrain when necessary to reach a goal. At the moment, NRP makes no attempt to do loop closure and therefore the network is always a tree of paths rooted at the ‘lander’. Figure 1 shows an example of a network of paths during one experiment. We have field tested NRP in three mission concepts. We found NRP to be a useful capability in two lunar-samplereturn scenarios carried out at the Sudbury (robot only) and Mistastin (robot and astronauts) impact structures [3]. We also showed how NRP could be used to conduct a Marsmethane-hunting scenario at the Canadian Space Agency’s Mars Emulation Terrain [4]. In the Sudbury lunar-sample-return scenario, a rover was used to carry out a geological investigation of a site. A remote backroom team (scientists and engineers) commanded the rover every two hours (for two weeks) by issuing goal locations. The rover drove autonomously to these goal locations using the NRP framework. The backroom team quickly discovered that if they commanded the robot to revisit a place, it could get there fairly quickly, compared to driving into new terrain where more caution was required. The parallel site investigation concept naturally emerged at Left image Image de-warp and rectification Stereo matching Right image Image de-warp and rectification Nonlinear numerical solution Keypoint detection Keypoint detection Keypoint tracking Pose estimate Outlier rejection Previous frame Sun sensor, inclinometer