Background
Over the past decade, SwRI has developed a strong reputation for providing innovative solutions to challenging technical problems in the field of automated vehicles. Because of this success, SwRI regularly receives requests to carry out very ambitious automated vehicle projects. Although SwRI is fully capable of achieving project objectives from a technical perspective, the corresponding cost uncertainties have limited the number of opportunities that come to fruition. Among the largest factors affecting cost uncertainties are a lack of thorough testing and refinement of the existing core vehicle automation software and limited overall functionality of the software. Maturing and expanding this software would reduce the risk associated with estimating these projects. This project was originally conceived in three phases. Phase 1 consisted of two main tasks: the preparation of the basic shuttle software and the implementation of a dedicated “people mover” vehicle. Phases 2 and 3 focused on the deployment, operation, and experimentation of the automated shuttle system, as well as the data collection and software refinement and enhancement process.
Approach
This effort bridged SwRI’s various automated vehicle technology offerings on a single vehicle platform. In Phase 1, we created an initial shuttle on a SwRI-owned passenger sedan, installed SwRI’s commercial automated driving system, and trained a safety driver to operate the vehicle on SwRI’s campus. The safety driver performed mileage accumulation to collect data and to identify and measure the limitations of the current system, especially rare events not seen during typical brief capability tests. In Phase 2, we identified two capabilities missing from the automated driving system that were critical for a campus automated shuttle: high-reliability intersection negotiation and intelligent detection of and interaction with pedestrians. The shuttle-building task of Phase 1 and the software improvements implemented during Phase 2 were shaped by these identified needs. Phase 2 concluded in FY22, and Phase 3, which involved extensive shuttle operations and data collection around campus, took place in FY23.
Accomplishments
During FY23 the shuttle received significant upgrades to the general campus tour, automatic log processing, and operator interfaces. The shuttle also received the first regular deployment of a camera-based detector using a neural network on one of our automated vehicles. Additionally, we began testing the pedestrian detection at crosswalks and automatic intersection negotiation. The shuttle successfully completed a six-month operational period where the shuttle operated autonomously 81% of the time. The shuttle has also received over 50 tour requests from 8 out of the 11 technical divisions.