Develop 1) vision-based terrain classification and 2) robust path planning capabilities, which together provides safer and more efficient way to traverse rough planetary surfaces.
Background: Achieving the goals of the 2020 rover and a notional sample retrieval rover (SRR) missions are even more contingent on mobility than MSL. Besides a very few exceptions, Curiosity’s driving speed has been at most ~70 m/sol, while the 2020 rover is expected to achieve an average of ~150 m/sol in order to visit geographically dispersed regions of interest (ROIs) and complete the sample collection. An SRR would fail if the rover cannot cover the distance to reach the cached samples. Furthermore, since MAVEN and ESA’s TGO are not in a Sun synchronous orbit, after the end of life of MRO the rover operations would become irregular, meaning that multi-sol drive would be required more frequently. However, due to safety concerns, the use of Autonav is currently limited within easy (e.g., “green” and “good blue”) terrains. Significant improvement in the reliability and efficiency of the Autonav capability is essential to future rover missions. Objective: 1) Develop on-board capabilities that enable rovers to traverse difficult terrains autonomously, efficiently, and safely. 2) Perform extensive V&V on test rovers and advance the technology to TRL 5-6. Approach: The proposed capability consists of two technologies: vision-based terrain classifier and risk-aware path planner. 1) The terrain classifier visually identifies terrain types (i.e., sand, bedrock, pointy rock) from on-board camera images. We employ a machine learning approach, where the classifier is trained a priori by labeled images generated by human experts. The learning is performed by using the Random Forest algorithm. 2) The risk-aware path planner computes a safe and efficient path by combining the results from the terrain classification and the geometric obstacle identification that already exists on Mars rovers. The planner is implemented by using a variant of the informative RRT* algorithm. Combining the two technology enables human-like reasoning on risk, such as avoiding >30 deg sandy slope or pointy rocks on a hard terrain, hence allowing rovers to safely traverse difficult terrains where the current Autonav cannot. We will deploy the algorithms on test rovers (Athena and ATRV Jr.) and perform experiments in the Mars Yard.More »
This work provides safer and more efficient way for the surface operation of the rovers in MER, MSL, and M2020 missions
Upon the successful demonstration of the proposed capabilities, they could be readily infused to future Mars rover missions including a notional SRL, as well as to future surface missions to the Moon, small bodies, and icy moons of gas giants. Also would be applicable to manned rovers.
Autonomous driving of commercial vehicles
Autonomous driving of military vehicles.More »
|Organizations Performing Work||Role||Type||Location|
|Jet Propulsion Laboratory (JPL)||Lead Organization||NASA Center||Pasadena, CA|