{"id":150,"date":"2020-02-19T21:06:55","date_gmt":"2020-02-19T21:06:55","guid":{"rendered":"http:\/\/mrsdprojects.ri.cmu.edu\/2020teamd\/?p=150"},"modified":"2020-10-10T02:24:04","modified_gmt":"2020-10-10T02:24:04","slug":"simulation-subsystem","status":"publish","type":"post","link":"https:\/\/mrsdprojects.ri.cmu.edu\/2020teamd\/2020\/02\/19\/simulation-subsystem\/","title":{"rendered":"Simulation Subsystem"},"content":{"rendered":"

The system needs to simulate the environment and external agent behaviors realistically. This<\/span>
is an important requirement if the system needs to work in real-world scenarios. Therefore,<\/span>
CARLA was the best choice, one of the most realistic simulators out there. Additionally, it also<\/span>
has a good developer community for support.<\/span><\/p>\n

CARLA can provide simulated sensor data streams of cameras, lidars, and radars. The sensors<\/span>
can be virtually placed anywhere in the environment or agent. This sensor placement will be<\/span>
made realistically which can be replicated in the real world later. Additionally, in order to<\/span>
accelerate development, we plan on initially getting data abstractions like segmentation maps<\/span>
and bounding boxes directly from the simulator.<\/span><\/p>\n

There are two main inputs to this subsystem one from the user and one from the system. The<\/span>
user inputs the types of behaviors of traffic that need to be generated. The simulation<\/span>
subsystem must produce a diverse set of scenarios in order to ensure our RL agent doesn\u2019t<\/span>
overfit to just a few scenarios. Furthermore, the system provides continuous trajectories<\/span>
that need to be tracked by the agent. The input trajectory is tracked by the simulator and<\/span>
updated sensor data streams produced from the updated environment will be sent to the next<\/span>
phase.<\/span><\/p>\n

\"\"<\/span><\/p>\n

Figure: Camera Stream\u00a0in Carla Simulation environment<\/span><\/p>\n

\u00a0<\/p>\n

Pedestrian Modelling<\/span><\/h2>\n

\"\"<\/span><\/p>\n

Figure: Modelling pedestrian crossing behaviour in CARLA<\/span><\/p>\n

We are also utilizing CARLA to train pedestrian detection and avoidance to our RL agent. We spawn a pedestrian in simulation and train our agent to avoid colliding with it while it crosses the road.<\/span><\/p>\n

\u00a0<\/p>\n

Improving Simulation Performance:<\/span><\/h2>\n

The \u2018performance\u2019 of CARLA is defined as the maximum simulated frames per second (fps) that can be obtained from CARLA during training runs, while minimizing the impact on computing resources so they may be used by other processes. There are several ways CARLA\u2019s actual and perceived performance can be improved that are listed below.\u00a0<\/span><\/p>\n