Problem Description

With increasing interest in autonomous vehicles in the industry, testing these vehicles becomes more and more important.
Autonomous Vehicles involve a number of subsystems. Primarily, the components can be grouped into Software and Hardware.

The hardware includes the sensors, the wiring, the embedded controllers, and the vehicle itself.

The software includes the controls software, the user interface, and most importantly for an autonomous vehicle – the autonomous driving software (includes the perception, planning, and the decision-making software).

The autonomous driving software needs to be extensively tested to ensure that it addresses the majority of scenarios that any vehicles would encounter. Testing in the real world is difficult. The edge cases that need to be tested include cases where the people or other vehicles in the environment are behaving abnormally. Also in the real world, the safety driver has to stay alert and engage the steering wheel well even in the slightest chance of a calamity. This usually happens much earlier hence the actual “edge case” might not even occur. Simulation helps solve this. But the problem with simulation models is that the behavior of the other people and cars in the simulation is not realistic enough. This is where our product comes in.

Our project aims to build a system where we have realistic models for simulation which have been trained on real-world data. We collect data from the real world and process it to extract the trajectories and behavior of the agents, which is used to build the simulation models.

S.T.A.R.S is an attempt to decouple testing and data collection, thereby allowing self-driving cars to learn from an abundant pool of real-world data at a fraction of the cost and more importantly without putting human lives at stake. It can also speed-up the iterative process of testing algorithm updates. To accomplish this, we will collect traffic data and build a model to learn realistic behavior as exhibited by human drivers from the data. The model will then be used to control cars inside a simulator.For this project we want our scenarios to reflect what happens in the real world and will,therefore, be a blend of good and bad traffic practices that self-driving cars need to account for.When we say ‘behavior’, it only refers to the motion of a car as a response to other cars,road boundaries, and traffic signs. We are not trying to model or simulate any other human traits.