3 ms·
What makes this really interesting is that all this is achieved with very little or no "AI" (in the context of current machine learning approaches). Purely heur
by freediver 5y ago
What makes this really interesting is that all this is achieved with very little or no "AI" (in the context of current machine learning approaches). Purely heuristics based approach to something as complex.
- ClumsyPilot 5y agoThere is weired tendency to use AI for well-understood problems that have formulas and solutions.
- habitue 5y agoRight, but is parkour a well-understood problem with formulas and solutions? I think there's also a tendency to go "You just need linear regression" a lot when actually, well, there is a reason people are excited about new AI techniques: they make things that were very difficult much easier, and they make some things possible that never were before. Is deep learning / reinforcement learning actually useful here? Maybe. I would be surprised if Boston Dynamics wasn't at least looking into it.
- gsibble 5y agoI just tried raising money. Mentioning ML makes investors immediately want to throw in. So hot right now.
- candiodari 5y agoWell. yes. Using AI means you learn one theory, and now you can do voice recognition, chess and parkour with state of the art performance (except perhaps when it comes to minimising compute) ... And then there's where it takes off. There isn't a single human that can come anywhere close to the language knowledge base AI algorithms can build up. Even a team of 100 humans would not have the breadth of languages algorithms have these days. Yes, context understanding humans still have the edge, but it's shrinking every day. But so many language aspects, from translation to grammar, machines outperform all but the very best humans, and 99.9% of all humans even within one language.
- auxym 5y agoWell, heuristics, and probably a lot of classical controls theory and inverse kinematics/path planning.