波士顿动力传授Atlas机器人新技能

Boston Dynamics Teaches New Skills to the Atlas Robot

波士顿动力（Boston Dynamics）尽管即将迎来新东家，但公司团队并没有停止传授 Atlas 新的技能。在官方近日放出的演示视频中，两台 Atlas 展示了接近于人类的跑酷能力。在 90 秒左右的视频中，展示了 Altas 跳跃、后空翻等，而且其中间杂着慢跑和突然的转弯。

Despite being on the verge of welcoming a new owner, the team at Boston Dynamics has not stopped teaching Atlas new skills. In a recently released demonstration video, two Atlas robots showcased parkour abilities approaching human-like levels. The roughly 90-second video displayed Atlas performing jumps, backflips, and even interspersed jogging and sudden turns.

更重要的是，这些跑酷能力并非一开始预先编程的。虽然对于波士顿动力来说，如果只是预先编程可能更容易做到。但当机器人必须处理多变的现实世界时，这种预设模式并不真正有用。为此，开发人员交给 Atlas 各种不同核心动作，然后机器人决定根据地形来组合和实施这些动作。

More importantly, these parkour abilities were not pre-programmed at the beginning. Although it might be easier for Boston Dynamics to just pre-program them. But when the robot has to deal with the ever-changing real world, this pre-set mode is not really useful. To this end, the Atlas was taught various different core movements by developers, and then the it decides to combine and execute these movements according to the terrain.

波士顿动力公司解释说：“在跑酷的这次迭代中，机器人正在根据它所看到的情况调整其剧目中的行为。这意味着工程师们不需要为机器人可能遇到的所有可能的平台和缝隙预先编制跳跃动作。相反，该团队创建了数量较少的模板行为，可以与环境相匹配并在线执行”。

Boston Dynamics explained, "In this iteration of parkour, the robot is adjusting the behaviors in its repertoire according to what it sees. This means that engineers don't need to pre-program the jumping actions for all possible platforms and gaps that the robot might encounter. Instead, the team created a relatively small number of template behaviors that can be matched to the environment and executed online."

尽管在仓库中后空翻或在办公桌上跨栏可能不是特别重要的技能，但机器人从一种行为到另一种行为并保持平衡和有效的能力肯定是重要的。由此产生的算法--由强度与重量比、运动范围、甚至身体健壮性等因素形成和制约--对波士顿动力公司的商业机器人，如Spot机器人狗，有更广泛的应用。

While performing backflips in a warehouse or hurdling over desks may not be particularly critical skills, the robot’s ability to transition between behaviors4 while maintaining balance and efficiency is undoubtedly important. The resulting algorithms—shaped and constrained by factors such as strength-to-weight ratio, range of motion, and even physical robustness—have broader applications for Boston Dynamics’ commercial robots, such as the Spot robot dog.

赞助本站

Sponsor this site