Field Robotics - Design & Development

Developing autonomous off-road systems requires the collaboration of multi-disciplinary teams working on different technology domains such as controls, perception, and motion planning. To facilitate this complex process while reducing the risks and costs associated with hardware testing, simulations are critical.

This talk will showcase various scenarios from a construction site to an open pit mine, demonstrating modeling and simulation workflows for autonomous off-road vehicles. Specifically, we will discuss the testing and optimization of autonomous algorithms through co-simulation using robotics simulators and popular game engines. Attendees will gain insight into the different phases involved in an autonomous offroad vehicle development project and understand how an integrated simulation workflow can enhance cross-team collaboration.

Building robots to operate in unstructured, off-road environments presents challenges that differ from those facing indoor mobile robots and autonomous on-road vehicles. Off-road robots must be rugged enough to stand up to both their outdoor environments and the unique tasks they are executing.

Sensors must work in all possible operating conditions. Software processing must be tailored to the task at hand. And off-road applications – like cutting thick grass, harvesting, or bulldozing – can be incredibly power intensive, which generally forces power savings in all other possible areas.

Using lessons learned from developing and deploying Scythe’s M.52 all-electric, fully autonomous commercial lawn mower, this talk will share the critical ways that unstructured environments influence hardware, perception, and software design decisions for off-road robots.

Building and testing field robots is challenging. Project timelines, tuning issues, budgetary restraints, and even crashes (both literal and figurative) can cause setbacks. There are a number of tactics robotics developers can use to reduce the risk of an expensive and/or dangerous failure, including breaking out individual subcomponents for testing and electrical or software integration, building standalone prototypes of subsystems, using a smaller, lower-cost robot to act as a surrogate for a full-scale system and simulation. This talk will explore some of these development methodologies, when they are appropriate, and how to employ them.

Robot sensor calibration is a frustration that can limit your commercial deployments. Calibration typically needs to happen in a controlled environment, with checkerboards, fiducials, or targets, and engineers present to run the cal process.

This talk will dive into sensor calibration in unstructured environments that don’t have checkerboards or potentially even any engineers. It will also explain how to achieve highly-accurate 6 degree-of-freedom pose estimates with or without GPS.