MER AutoNav and GESTALT

What AutoNav is

AutoNav is the autonomous navigation mode used by the Mars Exploration Rovers (MER). In this mode the vehicles use cameras to look ahead and avoid problems, combining stereo vision with GESTALT, short for Grid-based Estimation of Surface Traversability Applied to Local Terrain.^[1] The acronym names the job: estimate, on a local grid, which patches of ground the rover can traverse. The rover is still given a goal, but it does not wait for a person to select every steering command or every short segment of the route. It carries a model of the nearby ground and uses it to choose a path.

How a drive is built

A NASA/JPL description of rover driving says the software generates 3-D terrain maps from stereo images, considers dozens of possible paths, and then drives 0.5 to 2 meters closer to its goal before repeating the process.^[2] This is a repeated loop rather than one large plan executed blindly: look, build or update the local map, score candidate paths, move a short distance, and look again.

The scale of onboard terrain analysis was far greater than on Sojourner. Spirit and Opportunity typically measured more than 16,000 points from each pair of images, while Sojourner's onboard safety system could measure only 20 points at each step.^[2] A denser set of measurements gives the software a finer description of rocks, slopes, and other hazards in the local terrain.

Visual Odometry adds motion estimation. It is the autonomous detection and tracking of features in nearby terrain to measure how much the rover has slipped; the MER rovers were the first to demonstrate it.^[1] Wheel rotation alone can overstate progress on loose or sloping ground, so tracking the apparent movement of terrain features provides a check on actual motion.

Path planning has a local and a goal-directed part. In the revised AutoNav, Field D* evaluates the cost of travelling from the end of each GESTALT arc to the goal. Field D* was developed for MER by Anthony Stentz and David Ferguson of Carnegie Mellon University.^[3] GESTALT supplies local traversability options; Field D* judges how promising each option is for reaching the goal. The result is a short, safe move that is consistent with both the ground in front of the rover and the direction it needs to go.

What it achieved on the rovers

The effect showed up in time and terrain. Spirit reached the Columbia Hills 50 percent faster thanks to AutoNav, and both rovers could drive on slopes from 15 to 30 degrees thanks to Visual Odometry.^[1] Opportunity set a longest single-sol autonomous drive of 109 meters on sol 384 in February 2005, and a longest multi-sol autonomous drive of 270 meters from sol 383 to 385.^[1] A sol is a Martian day.

Version 9.2 of the flight software added four new technologies: the Field D* Global Planner, demonstrated on Opportunity; Visual Target Tracking; autonomous instrument arm placement; and AEGIS, for prioritizing science imagery onboard.^[1] These capabilities go beyond hazard avoidance. They let the rover track a target, place an instrument, and decide which images deserve attention, extending onboard autonomy from driving toward observation.

Where it sits

AutoNav is best understood as a set of capabilities rather than one program. It joins perception, local hazard avoidance, path planning, motion estimation, and, in later software, target tracking and science prioritization. The comparison with Sojourner's 20 measured points per step shows the change in onboard computation and sensing described by NASA/JPL.^[2] The MER rovers' first demonstration of Visual Odometry, and the use of AutoNav and Field D*, made the software a reference point for autonomous planetary driving: the rover observes, plans a short move, drives it, and then repeats the loop.^[1][3]