Visual Teach and Repeat (VT&R) is an effective method to enable a vehicle to repeat any previously driven route using just a visual sensor and without a global positioning system. However, one of the major challenges in recognizing previously visited locations is lighting change, as this can drastically alter the appearance of the scene. In an effort to achieve lighting invariance, this paper details the design of a VT&R system that uses a laser scanner as the primary sensor. Unlike a traditional scan‐matching approach, we apply appearance‐based computer vision techniques to laser intensity images for motion estimation, providing us the benefit of lighting invariance. Field tests were conducted in an outdoor, planetary analogue environment, over an entire diurnal cycle, repeating a 1.1 km route more than 10 times with an autonomy rate of 99.7% by distance. We describe, in detail, our experimental setup and results, as well as how we address the various off‐nominal scenarios related to feature‐poor environments, hardware failures, and estimation drift. An analysis on motion distortion and a comparison with a stereo‐based system is also presented. We show that even without motion compensation, our system is robust enough to repeat long‐range routes accurately and reliably. © 2012 Wiley Periodicals, Inc.

No abstract

Growing a network of reusable paths is a novel approach to navigation that allows a mobile robot to autonomously seek distant goals in unmapped, GPS‐denied environments, which may make it particularly well‐suited to rovers used for planetary exploration. A network of reusable paths is an extension to visual‐teach‐and‐repeat systems; instead of a simple chain of poses, there is an arbitrary network. This allows the robot to return to any pose it has previously visited, and it lets a robot plan to reuse previous paths. This paradigm results in closer goal acquisition (through reduced localization error) and a more robust approach to exploration with a mobile robot. It also allows a rover to return a sample to an ascent vehicle with a single command. We show that our network‐of‐reusable‐paths approach is a physical embodiment of the popular rapidly exploring random tree (RRT) planner. Simulation results are presented along with the results from two different robotic test systems. These test systems drove over 14 km in planetary analog environments.

A Mission Control Architecture is presented for a Robotic Lunar Sample Return Mission which builds upon the experience of the landed missions of the NASA Mars Exploration Program. This architecture consists of four separate processes working in parallel at Mission Control and achieving buy-in for plans sequentially instead of simultaneously from all members of the team. These four processes were: Science Processing, Science Interpretation, Planning and Mission Evaluation. Science Processing was responsible for creating products from data downlinked from the field and is organized by instrument. Science Interpretation was responsible for determining whether or not science goals are being met and what measurements need to be taken to satisfy these goals. The Planning process, responsible for scheduling and sequencing observations, and the Evaluation process that fostered inter-process communications, reporting and documentation assisted these processes. This organization is advantageous for its flexibility as shown by the ability of the structure to produce plans for the rover every two hours, for the rapidity with which Mission Control team members may be trained and for the relatively small size of each individual team. This architecture was tested in an analogue mission to the Sudbury impact structure from June 6-17, 2011. A rover was used which was capable of developing a network of locations that could be revisited using a teach and repeat method. This allowed the science team to process several different outcrops in parallel, downselecting at each stage to ensure that the samples selected for caching were the most representative of the site. Over the course of 10 days, 18 rock samples were collected from 5 different outcrops, 182 individual field activities -such as roving or acquiring an image mosaic or other data product -were completed within 43 command cycles, and the rover travelled over 2,200 m. Data transfer from communications passes were filled to 74%. Sample triage was simulated to allow down-selection to 1kg of material for return to Earth.

No abstract