This live stream webinar highlights physics-based real-time radar simulation applied to a reinforcement learning model for performing longitudinal vehicle control.
The Real Time Radar (RTR) simulation introduces a new paradigm for sensor development by leveraging GPU acceleration and new algorithms to accelerate simulation by orders of magnitude without compromising accuracy.
We will demonstrate how to use the open source driving simulator CARLA and RTR technology integrated via the standardized open simulation interface (OSI).
Discover how RTR addresses key ADAS and autonomy use cases including:
- Many targets in the radar field of view
- Multipath and interference among complex scatterers
- Consideration of widely varying surface scattering properties
- The micro-Doppler impact of motion in the actors’ local frame of reference
- Multiple-Input Multiple-Output (MIMO) radar.
Additionally, we will demonstrate the automated development of a longitudinal controller that regulates a vehicle’s cruise velocity based on radar sensor returns.
Lastly, a reinforcement learning model will be trained to consume raw simulated sensor data and successfully control the longitudinal speed of the vehicle. The model will learn incrementally from its mistakes while subject to a massive number of scenarios. The progress of the reinforcement learning will be streamed at two-hour intervals over a period of four days. By observing the changes at each interval, you will see how the learning module incorporates new data and improves its ability to control the speed of the vehicle over time.
Please register for each day of this event separately.