Speedway 14.69 seconds

[Lotharsson]

Is it just me or is PSO useless for full distance tracks? How does time factor into the distance measure?

I don't believe I've ever had it help for tracks that run to completion. It seems to devolve, either to a whole bunch of cars that rapidly fail, or to a bunch of cars running slower than the input seed. Perhaps it will help if the input seed isn't as highly optimised though.

[Lotharsson]

Looks like 14:73 and 14:69 are magic numbers somehow...

It was quite difficult to get a 14.66, and a lot easier to get 14.73s and a few 14.69s. While I may return to it later I am currently skeptical of improving on 14.66 and I'm looking at other tracks, so here is my car:

Code: Select all

eNpz8FxfXqUbJuvAwQAGDtj52jC+/c27VxKD1rk4MHY/veBz5jhUnp+AfnQ+I4wPBCz2j79sY7wQtcD+B1yM7QDHsn0/9hStQxJjdGAV1l38JOeq/Yv3mydFzpwIUrf/c+a7zQdX77V/+96/bVlaMlgvS8fn66orl9rfepfZs9lYGazuV5i6wOZVYfbP+F6KSM6cBLb3vyXznpy1QvbfK4/F2P+/BRRj2v/BR1Hee4eX/au400vS0uIZnsdXMZRcusbQVtzIYMyQwLDebjXD1uo8himTuxlKj59lSL1lxbCBgR8srnfxN0OVUDXDx9LLDAKcMxmMLs0AmQkAo9904w==

I ran a bunch of GA and PSO on it without getting any improvement. Maybe it will trigger someone else to have more luck.

[Diseased]

Is it just me or is PSO useless for full distance tracks? How does time factor into the distance measure?

I don't believe I've ever had it help for tracks that run to completion. It seems to devolve, either to a whole bunch of cars that rapidly fail, or to a bunch of cars running slower than the input seed. Perhaps it will help if the input seed isn't as highly optimised though.

I've found that running PSO with highly optimized cars results in a fairly long period of genetic slush cars before it re-converges on an approximation of the original car design. Does help in the end.

[ags]

... fairly long period of genetic slush cars ...

How long are we talking about? I gave up after 2000+ generations with default parameters. It seemed to be converging to a silly 16 second car.

[Diseased]

... fairly long period of genetic slush cars ...

How long are we talking about? I gave up after 2000+ generations with default parameters. It seemed to be converging to a silly 16 second car.

Wow! That is quite crazy, in my experience I would say that at the low hundreds of generations at worst it re-converges with good results. I've seen it happen on Speedway, Sisyphus and Tight Squeeze.

[Lotharsson]

I've also tried 2000+ generations several times with what appeared to be convergence that ran nowhere near the completion speed of the input seed.

I guess you could say YMMV...

[websorcerer]

Here's another one. I started out with a completely maxed out octagon with wheels at 4 and 6. I then minimized the magnitude of 2. This is the result of overnight optimizing:

Code: Select all

eNqzf6JyV19q/n4HDgYwsH/0dmJG0QZFOP9V4K/jIPry4pQpOmfO22/0DI8vFBTCqf7R/H1lJ7ZYIfiBr1vldgThkgcCFvs3avz2K1ry7H/AxdgcuFU2u7UL/oGJ/QcC++8bs7gXSn6zfxH9fFbPmTNAYVYHXrly3+5nS+yv3p/+xAcsxuzAduCJ/eSTf+0fLxc5kKSkCBRjsv/4Q+F3EJOV/RXWZ/rby8tB5jmwJd39cCGl3/641r05aZ1vwXo530VbrfjsYP/MKOQsyLz5DC8Zngo/ZDj3rYkhYsohBhW1dhR8Zosug8pSSTD7k7Ivg0VYK0PIAgEGxS86YDGg3QBJOnwR

[Lotharsson]

Here's another one. I started out with a completely maxed out octagon with wheels at 4 and 6. I then minimized the magnitude of 2. This is the result of overnight optimizing...

Very good - next step, get down to 14.66 I found a second one that does, but no 14.63 yet:

Code: Select all

eNqz/8AABg7s/zmnrvo+wR7G54DQGPwGmYepzgr/HRi7n17wOXPW/v1/EHiDUz0BPhCw2D/+so3xQtQC+x9wMTYHzkUrA648M0ESY3RgFdZd/CTnqv2L95snRc6cCFYn4l6/PPiQpv3b9/5ty9KSwWK8u5kuJm3otr/1LrNns7EyWEyop3CCuFOT/TO+lyKSMyeB7f1vybwnZ62Q/ffKYzH2/28BxZgcRPk+8xTas9m/iju9JC0tnuF5fBVDyaVrDG3FjQwhxvwM6+1WM2ytzmOYMrmbofT4WYbUW1YMGxgg4noXfzNUCVUzfCy9zCDAOZPB6NIMkJkAGwxqxg==

[websorcerer]

I got one down to 14:66. I started with a maxed out octagon with wheels at 3 and 6. Made the bottom parallel and close to the ground. Shortened 4 to about 50% and optimized overnight at a mutation rate of 1.

Code: Select all

eNqzf5fN1as5c5YD+3/Oqau+T7D/wAAGDhwQmiD/dqvcjsDXrfafN+nyhbFI2j8/daBZKG+GA7tpe1/qzFkY6t+vSWS54+IC4wMBs/3jL9sYL0QtsP8BF2Nz4Fy0MuDKMxMkMUYHVmHdxU9yrtpfKfDftSwtDKxOxL1+efAhTfunkY8Kfc6cAovx7ma6mLSh2/7Wu8yezcbKQDFWB6GewgniTk32O2eCwCywvf8tmffkrBWy/9bCu/Bk7n+gGJODwPLZtxJZ7tq/iju9JC0tnmGOtScDm5wMQ2VaCcNWvQUMiwznMSQe+sbwW2Umw/0LWQwRZ3UZFjMwMfyY/YlhO+MUhlzWPAaXQjGGJ/m5DPm7u0BmAgCuP26T

[Lotharsson]

I got one down to 14:66. I started with a maxed out octagon with wheels at 3 and 6. Made the bottom parallel and close to the ground. Shortened 4 to about 50% and optimized overnight at a mutation rate of 1.

Excellent! That's a different approach from mine. Wonder if you can get one to go faster still?