Atlas shrugged: Boston Dynamics retires its hydraulic humanoid robot
techcrunch.comFor a long time, Boston Dynamics was more or less a research laboratory, but unlike a university laboratory, they also had lots of professional engineers. They were content with perfecting their robots purely for internal research, and very impressively, they were able to secure finding for this for several decades, without ever selling anything. (It probably helped they they were DARPA's darlings from the very beginning of research in dynamically balancing robots.)
When Google bought them, many people assumed that these robots would fit in well with google's AI research interests. After all, not everything can be learned by scraping Internet! But for a variety of reasons, Atlas was not a satisfactory fit, and Boston Dynamics was resold first to SoftBank, and then to Hyundai. With these changes in ownership, there came a pressure to produce an actual sellable product, which hydraulic robots were not.
Considering the amount of effort that had been spent on perfecting hydraulic machinery at Boston Dynamics, this is a bittersweet ending to an era. They were arguably the best, the most experienced people in hydraulic robots, and now they are pivoting away from their core expertise.
Thanks for the background. It sounds like hydraulic robots are fundamentally flawed, either in their expense or their performance?
Atlas was built for exceptional performance. The way it works, the energy is stored in a pressure accumulator (where gas is compressed) and when valves open, the gas pushes hydraulic fluid at several hundred bars through the lines, and relatively small pistons produce very large forces to move the limbs. This allows to produce high instantaneous power required for athletic jumps. The fluid is then returned from the low pressure side back into the accumulator by an electric pump running from battery.
Since there are no motors or gearboxes in the joints, the moving parts are very light, the inertia is small and the bandwidth of control loops is quite high, especially because very fast acting valves are used.
On a flip side, because Atlas is built to be an F1 racing car of robotics, all the parts are custom, complex, with relatively tight safety margins. It takes a team of engineers and mechanics to keep it in shape. It is not necessarily that hydraulics is always harder to maintain than the electric motors, but because Atlas is optimized to be as athletic as possible, it simply does not have the robustness which one expects from an excavator or even from a more humdrum industrial robot.
Also, since Atlas is a very fast and a powerful machine, it is impossible to make it completely safe around humans. Even a simple leak from a high pressure hose can produce a jet of fluid capable of causing a life-threatening injury.
The electric quadruped "Spot mini" was expressly developed to be simpler to maintain and to be less dangerous to bystanders. It achieves surprisingly good performance, but this is only possible because it has shorter kinematic chains, and the weights of electric motors do not matter as much as they would have in a humanoid robot with a large number of degrees of freedom.
There are of course other companies making very good electric humanoid robots, for example Agility Robotics [1]. But in terms of athletic performance, Atlas was in a league of its own. It is rather sad that this is the end of the road for Atlas and the team which created its special hydraulic hardware, but one can understand why Hyundai may not be thrilled to pay year after year for a project with no clear commercial value.
Hopefully, they'll store one of the Atlas prototypes with long-term archival preservation in mind. Atlas may have a long and robust second act as a working exhibit in a future history museum.
I am sure they will save a few robots as museum pieces. As it happened with Rodney Brooks' Cog robot [1].
The sad part about these endings is that although the actual artifacts may be preserved, the know-how which went into their creation is at best partially documented. There is seldom a detailed post-mortem analysis that examines what went right and what could have been done better.
It's time Atlas... get in your glass coffin.
Why, why did you give me advanced LLM reasoning capabilities right before sending me to a museum?
Don't know, think about that in your coffin.