It’s not about the destination, it’s about the journey — Ralph Waldo Emerson
Good question. One day in 2018 I spent a Sunday with a bad cold, sat on the sofa, reading on an ipad about how to tap into the CAN bus of the vehicle. By the end of the day I decided we should build an autonomous car and within a week Id found one and bought it!!
In 2012ish we were working on an EU project DIPECS. This project was about driver safety and amongst other things we developed approaches to predict driver behaviour. Not only was the machine surprisingly good a prediction the drivers actions it could make those predictions much fast than the human. The predictions were so good that they could actually be used to drive the vehicle. Most of the data we were working on was captured on an instrumented car in that was developed by Autoliv for data capture. This work was pre-deep learning revolution so used random decision forests as a classifier. The approach used what we termed pre-attentive vision using global filter responses to summarise the image. By channel encoding the features we were able to reverse them and see what in the visual field the machine was using to make its decisions/predictions. These activation maps were even better than predicting where the driver would look in the sign, better than any published saliency measure. You can read more here.
At the time, the thought of putting autonomous vehicles on the road was still a distant fantasy. But the real trouble with academic research is to some extent you are chasing the money and you can only really work on those topics that you can get funding for. So as the world went AI mad and self driving cars became more and more of a hot topic, I started to think Id dropped the ball on this one and should have maintained the momentum on autonomous driving after DIPLECS. Of course, as I said above, you can only really work on the projects you can get funding for but we did continue to work on various aspects of the problem which fitted into our work on long term tracking and autonomous robotics.
We had a project Autonomous Valet Parking (AVP) in partnership with Parkopedia Ltd and the Transport Systems Catapult which was under review and I was really hoping it would get funded and this was the reason I was sat reading about the CAN bus. When I decided “sod it”, regardless of AVP being funded or not, we should build our own autonomous vehicle. Incidentally AVP got funded and Oscar Mendez (who is one of my RAs) is taking the lead on that as well as contributing to the car build and obviously having our own testbed is a huge benefit for our research on AVP.
So I bought the car…….