One of the greatest challenges many artificial intelligence projects face (specifically deep learning related projects) is the high costs associated with hardware. Consider, for example, Google’s self-driving car: this car makes use of the expensive LIDAR radar system, and as such is an unrealistic option for many private organizations.
In our research, we will try to create a self-driving vehicle using minimal hardware, shifting the focus from sophisticated hardware to an algorithmic approach. This has a number of advantages:
- Costs go down significantly (from millions to thousands)
- The system learns how to function in a more abstract environment, which means the algorithms may be reused for other types of vehicles
Using a new learning method in the field of deep learning, we aim to create an A.I. that pilots a boat using only the feed of five affordable cameras. In a later stage, the system could be trained to drive other vehicles as well.