Autonomous Dynamically Simulated Creatures for Virtual Environments: Animations & Screenshots

Autonomous Dynamically Simulated Creatures for Virtual Environments

Paul Urban, Computer Science Honours Research Project, 2001

Animations

One of the creatures built was a roobot, a hopping planar monopod. Its hop is controlled by a finite state machine with 14 parameters. A genetic algorithm was used to evolve fast and stable hopping.

The animated gifs may take a while to load...

Generation 0, Fitness 72.83
The initial (guessed) parameters - the roobot falls over after its second landing.

Generation 92, Fitness 341.01
A fairly stable parameter set after almost 100 generations, but not very smooth - it hesitates after each hop.

Generation 602, Fitness 454.64
After several hundred generations the motion is almost completely smooth, and getting faster.

Generation 8927, Fitness 887.43
After close to 9000 generations the motion is completely smooth and very fast. Even though its tail sometimes touches the floor, the roobot maintains its balance.

Screenshots

Underwater environment with jellyfish

Back to my CS research page...

Permission to reproduce all or part of this work for non-comercial purposes is hereby granted, provided that the original source is referenced in full.

	Generation 0, Fitness 72.83 The initial (guessed) parameters - the roobot falls over after its second landing.
	Generation 92, Fitness 341.01 A fairly stable parameter set after almost 100 generations, but not very smooth - it hesitates after each hop.
	Generation 602, Fitness 454.64 After several hundred generations the motion is almost completely smooth, and getting faster.
	Generation 8927, Fitness 887.43 After close to 9000 generations the motion is completely smooth and very fast. Even though its tail sometimes touches the floor, the roobot maintains its balance.