Sisyphus: completed (maybe)

[fstuurman]

Sisyphus 250.9.png (9.01 KiB) Viewed 1160 times

eNqzP2Zsvz66hN+B/R9T4NXse/YrF0TY6JVoODDILZ6zrCDH/tkMlb733wLsLx7qeZpZHmj/gQEMHDggtP3KmSAwy4HRg/niHodPML791x6l7Y9fzYSpt98JFX+7Wklmy9kEB3bbNwmm/+P+A4GD2MurrDrdufavKj9uXe+YAzafs3L9iUqxHvu34nfufZLqBAoxO4iJNZW9znlj/3Lv+/fVu7WBYowOEh0Hl3G28Nh/CvVMXaQgAhKz/ykRGmHBvMD+04ri2cx3PoPtYHjg4X9qi6T9o1K/w7I3e0Bi9t/TrieH8irYf619tsS1YTJYHduSlc2bdz+1/67GuyHQXJShqvIcg+SJEgYeuxyGYltZhuJ1TxlWLqpiyOHhYZgrGchQmmvF0Cz5hoFn4TuGVD9PBk01HgbJd5IMb3kuMkyfbglyNwAhc5YJ

I worked several days on the assumptions
- mass should be as low as possible
- center point should be as low as possible
- mass should be pressing on front wheel, preventing it from lifting

Finally settled on manual design, a triangle between ground touching the wheels and the axis of the front wheel. First car created manually hit 250.6. After 30 generations PSO the 250.9 was hit.

[fstuurman]

The distance in Sisyphus is measured horizontally.
But if you consider height reached, look at this one:

Sisyphus high.png (8.49 KiB) Viewed 1161 times

eNqzP+1RszzGR9WB/Z977PXF5fYr53WUWM2zsv/uf9FowdZo+6d9qalc89jsD76+oz7bzNT+0Sy1ovvXX9mf+5Z+8emXbvun6hud+y8bOrD1HTjFfJPHfuVMEJjlwFzqtl5g7yn7DwxgYH/rSbzBTt3F9m8vsz5W9lzgwO7oWeSg2/gfCBzEWvwN5hgV2r/a/3dGVYclSL0Dx474LgcPefu3qSkMcn8FgUIsDqKfKqyyLu+0f7m/r+8h2w6gGKODRHuD9F+dV/YfWypOGQaYgcTsf+i2/l0cImr/ydSc7fKsHrAdjAILD8zc8NT+sXG7cIp2AUjM/oew8a5tNy7af1PL4P7yVAGsjr1rTWDAon3239POfiy+OZ1hyoS7DJI3axl4HHMZWpy0GaaeecawZX4JQx4PL8Me+VyGkgI3hlbJzww8Mz8xFMVlMmhb8TAo/FBm+MFzhWHePD+ge5gBJ9ecpg==

You can see the top! Only two segments to go.

[Lotharsson]

...But if you consider height reached, look at this one...You can see the top! Only two segments to go.

Excellent - as is the 250.9 which I haven't been able to reach yet!

One of my 250.8's from a few days ago has been able to see the top, although it doesn't get as high as yours and I haven't been able to find a way to keep it on the track that little bit longer yet:

Sisyphus 250.8 track end.jpg (17.6 KiB) Viewed 1153 times

[Lotharsson]

I worked several days on the assumptions
- mass should be as low as possible
- center point should be as low as possible
- mass should be pressing on front wheel, preventing it from lifting

I've reached similar assumptions, except for the low mass. I can also reach 250.6 with hand designs that have significantly higher mass, but haven't yet seen them improve further via evolution. However the scores from hand designs that follow these principles vary all over the place. I think you need a bit of luck too - you want a smooth ride along the steep parts of the track, because once a back wheel comes off the track it can slow the car down too much to recover, and once the front wheel comes off the gravitational forces make it hard to get back on again. I don't know of any principles that let you design for a smooth ride.

[fstuurman]

The evolutionary goal is not a smooth ride. On the contrary, the selection will exploit any irregularities. And I think that these irregularities are highly dependent on very small differences in design. Also that small rounding off errors in the simulation get amplified to these irregularity's.
If my theory is correct that would explain performance differences in Mac and Windows versions. Just using a different compiler for the flash would create a different code. Which cause tiny differences which get amplified.

[Lotharsson]

The evolutionary goal is not a smooth ride.

No, but I believe a smooth ride near the top on Sisyphus is critical to the evolutionary goal, which is a very high score.

And I think that these irregularities are highly dependent on very small differences in design.

From my experiments changing one value in the chromosome by the smallest possible increment (<10^-15 in most cases) can be enough - at least on some tracks and designs - to take a top score down to mediocre (or vice versa).

[Metaldog]

eNqz3/00hnHeckEHtmeVp67M67bfaWw0LYq52IFNsu755fQQ+wO2SR+evcq1//ruV+ze9in2R9/vsuqye+vAMlVLp136iv0FOVuhzL6n9h+nf+H/qeVo/+78c9+1P0od2DsFV4g7qti/f5UxKVmh2X4PYxhPtMQ0+w8MYODAXsra4B8UDGSyO/C0ti55UOBl/6Y6xDBgZTdQjNVB+B1TXoR/k/0rSZ/SqNU6/4HA/oNRekiK0Uv7j4GT2zatDQCLvWtz284lesb+lb29vr7LP7DZ/DMZOMwu6dh/9yq8zzq3GGye6GfBVzMWW9k/WJjGWNuXBFJn///RKv03i2fZf/21qCjTcAXE3hvx6YFVvPbXVyzK21fzkIGHZzaDpORcID0VSE8F0lOA9HQofyaQngWk5zAI8txg+CM5j0E8cxJDh8FvoPhRhgJJGaCZTACfPIlP

doesn't quite reach the top but a piece falls off the track and somehow hits 251.3 on its way down

[Lotharsson]

doesn't quite reach the top but a piece falls off the track and somehow hits 251.3 on its way down

Cool - that's a much bigger piece than mine was (http://www.boxcar2d.com/board/phpBB3/viewtopic.php?f=5&t=43) and yours isn't using the negative mass bug either!

[fstuurman]

Made it to 251.1

Sisyphus 251.1.png (11.65 KiB) Viewed 1121 times

eNqzP3r+0K6LdfkO7H9tJ0c96rI/kXRV8ZTlLPv3G98+WmgZY/+ctS3+Ue9t+2sJi3R/lovZP+dceSmxbKcDBwMY2K+cCQKzHBglQlpq+L7A+PZf48M51vlPt/8AVbcTKv7GfW3kEfcQB7ZfrsE7vqz5DwQOwlWff2m5v7F/1cV/8uflFyD1DpwNT6pUpj2yf2u99vackGCgELODmA+f/Jrrt+1f7jd86dPaDBRjdJDQLnh4tMrc/tPUH2r/T7mCxdhq53217+ewf9fy6cm5NQZgOxh+/2r/wPXa/oW1v5a/SAxE7HYW0DEK9p8LI77HXogAi7FNEOteL/jF/rt2WFVH1T4G4Q03GSR/ajPI8tQw3F8gx6A/6SLD68/TGEx4eBjalewYXJ31GI46/GfgsVvIcFKykCEsiIdB8rkkwzGefQy1LkkgdwMAzHmezQ==

The PSO has enormous problems going forward now. Almost all variations scrape the floor.

[Lotharsson]

Made it to 251.1

Legend!

I tweaked one of yours last night and got the back wheel even higher - to the middle of the last black segment - but I couldn't get it to stay on the track any longer than the original did (250.7).