One of the most challenging problems for autonomous ground vehicles (or wheeled robots) is navigating in an urban environment. This is mostly due to the unpredictable nature of pedestrians and other vehicles. When coupled with the usual challenges of inaccurate and non-linear dynamics models, noisy sensors and finite compute power, solutions to tasks such as overtaking and crossing intersections are far from straightforward.
These problems are often addressed using more expensive hardware (high-resolution cameras, faster on-board computers, etc.), while sticking to standard techniques and algorithms. An alternative approach is to rather maximise the information that can be extracted from a more affordable but less powerful system, which requires careful design and integration. This project will investigate the trade-off between these approaches for a set of driving tasks in the standardised CARLA simulator.