Team Popcorn

Thursday, March 03, 2005

Physics model

Good news guys, I have finally finished putting together the physics model. It takes in the weight, linear velocities of the agent and the desired velocity and calculates the power required by the engine.
But I must add, there are quite alot of bits and pieces that needed adjusting, but before that I wanna c heck it with the agent.


Tuesday, February 22, 2005

An Update

The project is still coming on in leaps and bounds, we have set ourselves some objectives for the coming couple of weeks which will hopefully end up with most of the implementation completed.

Highlights in development..

  • Track Drawing in the Simulation, still need to add ability to draw track within application. (currently in separate program).
  • Tried out varying methods of training cars.
  • Had a go at training cars to go left, even though there parents could go right :)

Fixed this evening...

  • Fixed Screen Refresh Problem, whereby graphics of cars would get stuck on screen.
  • Added ability to Start/Stop Simulation using buttons.
  • Added ability to Save/Load Genetic Algorithm Populations (Cool feature!)

And a nice screen shot.. more to come

Slightly evolved cars attempting to go around a figure of eight track :)

Saturday, February 19, 2005

Whoa Neural Network Success!!

After several hours modifying the harness code and tweaking the Neural Network module, we have managed to get the cars to 'Learn' to navigate to a point.

After training a population of cars on a few points, they can then navigate to other points in the world on their own! Its a massive success for the project and another milestone reached.

Well done guys!

Thursday, February 17, 2005

Mid Feb Update

Just an update of the project progress..

  • We have managed to build a demo whereby a car navigates to a user set point using simple rules.
  • We have created a test harness for the Learning side of the project.
  • We have connected the Neural Network/G.A module into the harness for testing. We may possibly need to implement a Backpropagation Algorithm.
  • The network seems to only output limited values, so we will need to modify this module.
  • Got the cars drawing pretty behaviour patterns...

Thursday, February 10, 2005

Neural Net Update

No glamorous photo, but it seems the neural net is now working. It uses unsupervised learning, updating the weights through a GA. The next step is to plug it into a test harness, and see whether it can learn to drive a vehicle towards a checkpoint. Details to come...

Monday, February 07, 2005

Differential Steering Modelled!

After sometime scratching our heads and playing around with maths, we have been successful in modelling the Differential Steering required for the cars.
Screenshot below:

It should now be relatively straightforward to plugin the neural network to get some hopefully intelligent behaviour.

The next step is to get a single car to learn to navigate to a particular location within the world, and build in some basic collision detection on the bezier track.

The basic physics procedures are being coded, and after that getting it wrapped together into a 'physics model'.

We now need to look at each key system in the project and work out which tasks require attention, and set out a time scale for each task.

So overall we are making good progress..

Well done guys...

Tuesday, February 01, 2005

More Success on Bezier Curve!

After several hours scratching our heads, we have finally managed to get the track drawn using a bezier.
Now for the A.I :)
Screenshot below: