Thousands of people are about to start training behaviors on real, tiny humanoid robots
热度趋势
百分比基于当前可用热度信号，而非评论数或独立用户人数。
Pollen Robotics 的一位工程师透露，该公司已发布了一款新型机器人，预订量超出了所有预期，达到了每五秒钟一台的速度。这位工程师表示，希望人们能通过这款机器人获得乐趣、学习知识，并为社区做出贡献。该公司对听取社区对此的看法充满期待，尽管有关于英伟达收购他们的传闻，但他们更关注新产品的发布和用户参与。
Pollen Robotics engineer here. I’d like to discuss a deeper subject than the rumors about Nvidia acquiring us. Today we released a new robot, pre-orders are going crazy (1 robot every 5 seconds, beyond any expectation we had!).
Some will see Microduck as a toy (and that’s fine, go have fun!) but I think it’s important to see that this thing is a full-on biped humanoid robot with an open-source software stack and a sim2real pipeline that works surprisingly well.
This means that, in a few months, we’ll have thousands of enthusiasts trying their hand at reinforcement learning on a real robot and generating new behaviors. This is a complete shift from the LLM landscape where you have a few huge entities that train huge models.
I wish this existed when I started doing robotics in 2010. I expect the rate of improvement to accelerate. Especially with coding agents being so good that almost anyone can work on this.
I hope people will have fun with it, learn a lot, and contribute back a lot. Curious to hear what this community thinks about it!