When designing automated robotic things, such as self-driving car, training is almost everything. Yes, you have to train the car, or more precisely – train the algorithm that will be responsible for all driving related activities.
Image credit: MIT CSAIL (still image from YouTube video)
Here machine learning this process is quite difficult to do and also to ensure that this task is done both exactly and efficient. All training models and algorithms must be properly constructed. Furthermore, driving situations can be very diverse, and therefore it is very expensive and time consuming to put the developed prototype on a real road to gather all the necessary ‘experience’.
However, you can now facilitate this work with VISTA, a new open source emulator from MIT CSAIL released just a few days ago. Here is a short video showcasing its key capabilities:
The developers of the new tool describe it as:
VISTA is a realistic, data-driven simulator for autonomous driving. It can simulate not only live video but also LiDAR data and event cameras, and combine other simulation vehicles to simulate complex driving situations. VISTA is open source.
Technical paper: https://arxiv.org/pdf/2111.12083.pdf
Code: https://github.com/vista-simulator/vista
Document: https://vista.csail.mit.edu/
