This test scenario dictates that an autonomous LTV (the Polaris RZR PRO R 4 ULTIMATE, in our case), which is the system under test (SUT), continues to drive straight on a dirt road while continually performing visual servoing (VS) in order to execute a panic braking maneuver in case it encounters any animal(s) along its mission path.
To this end, we devised a perception module that makes use of AI-based object detection models to detect and classify objects in the environment (it is worth mentioning that these AI models are not particularly trained on data from off-road environments with objects such as moose and serve as mere candidates in this research). Furthermore, a planning strategy was formulated, which determines whether or not to trigger the autonomous emergency braking (AEB) functionality. It observes the class of the detected objects along with their sizes and classification confidence to analyze whether these objects actually exist (filter out false detections) and pose an immediate threat/liability to the vehicle (larger the object’s size for a given class, more proximal it is to the vehicle). Finally, the AEB functionally controls the vehicle’s throttle and brake to keep driving under nominal conditions and apply hard brakes in case a collision is imminent. Additionally, since this primary autonomy algorithm relies on visual perception, a secondary algorithm was devised to control the vehicle lights based on ambient light and fog/mist present in the environment. These values were inferred based on the time of day and weather conditions.
AutoDRIVE Ecosystem Website: autodrive-ecosystem.github.io
RZR Digital Twin: • Digital Twin of Polari...
Conference Paper Preprint: arxiv.org/abs/2405.04743
A similar video showcasing on-road vision-guided autonomous emergency braking (AEB) for OpenCAV digital twin is available here: • On-Road Vision-Guided ...
Негізгі бет Off-Road Vision-Guided Autonomous Emergency Braking for RZR Digital Twin | AutoDRIVE Simulator
Пікірлер