The Cavenwell AI learns the design of a car suspension
The Chrono physics simulator comes pre-packaged with a vehicle simulation, the HMMW vehicle. It was developed by a team at the University of Wisconsin-Madison. This car simulation includes a detailed model of the suspension, specifically a double-wishbone suspension. The simulator is capable of outputting the positions of all the parts of the model and this output can be massaged into a CSV of 600 columns, where each column is the position of different parts of the vehicle at every time step of the simulation.
Using a powerful, parallel cloud machine, and taking only 5 minutes, the Cavenwell AI was able to regress these 600 columns into an algebraic theory of the car suspension.
A car suspension is designed so that each wheel assembly is independent. This is to ensure the simple effect that if a wheel rolls over a bump, the rest of the car does not react. The AI correctly detects that each of the (left-right, front-back) assemblies is independent, and it does this by simply offering no equations that relate left-right-front-back. The AI instead produces equations showing how the parts of each individual assembly (e.g. the parts of the left front wheel assembly) are related by some equations. It finds equations that model the graph of interrelations that we see in the design of the suspension. The algebraic theory can then be used to do automatic reasoning about how the car works and the first non-trivial deduction was performed today.