drive-primer: A high-fidelity simulation suite for mobility planning

About this technology

Overview

Current mobility planning for downlink operations of the Mars 2020 rover depends on low fidelity slip tables parameterized by discrete terrain classes. Extrapolation of rover performance on these broad terrain classes is difficult and oftentimes inaccurate. This is especially crucial during mobility faults with highly non-linear and time-dependent dynamics on the wheels, where we can observe slip-sinkage, embedding, and cross slip. We propose drive-primer, a Bayesian optimization-driven framework to replicate mobility faults in a high-fidelity Project Chrono CRM/SPH simulator. The contributions of drive-primer are three-fold. (1) A parameter identification method to discover the local physics based parameters of terrain that has been driven over -- reconstructed in simulation with stereo-camera height maps, trajectory/poses generated by VO/SAPP, spacecraft EVRs, and joint telemetry. (2) This terrain parameterization is in a continuous parameter space (rather than broad classes), which is useful for qualitative comparison and analysis with respect to previous drive segments. (3) We demonstrate a use case, where with our dp-table tool and calibrated parameters, we generate a slip table that fuses telemetry and slip-slope CRM/SPH simulations into a Gaussian process for a more precise and 'high fidelity' slip table.