4 ms·
The metrics they're using (2x its height for climbing a wall, 2x its length for crossing a gap) are weird and don't really relate to the same achievements for a
by s3krit 3y ago
The metrics they're using (2x its height for climbing a wall, 2x its length for crossing a gap) are weird and don't really relate to the same achievements for a traceur. 2x its height is really more like slightly over 1x its usable body for that maneuver (0.4m length, 0.51cm height of the climb). I agree, not extreme but still pretty impressive for a robot. We're not going to see them doing cat leaps any time soon ;)
- GraceCat123 3y agoTotally agree with this! However, it's common for research papers to use exaggerated language to emphasize their breakthroughs and accomplishments. :)
- notahacker 3y agoRelative size of leaps doesn't seem to be a particularly useful metric for assessing the NN performance anyway, since in the absence of the human constraint of fear it's really limited only by the mechanics of its legs and its weight. More impressive would be adaptation to obstacles without clearly delineated edges, sticking landings on uneven/moving landing sites, and especially avoidance of landing sites which appear incapable of supporting its weight properly, particularly if it could do it well enough to generalise to novel courses. The video hints the model may be able to do this to some extent (the high jump does show some apparently necessary compensating movement to avoid slipping off), but doesn't really demonstrate it.
- chris-orgmenta 3y agoI'll extend 'avoidance of landing sites which appear incapable of supporting its weight properly' to 'avoidance of landing sites which are off-limits, including living things' E.g. A sleeping dog that is motionless, a valuable item, wet concrete. I suppose I'm going out-of-scope, though.
- notahacker 3y agoI'd watch the video of one of those stepping on a sleeping dog tbf...