It's outright terrorizing to imagine what a weaponized TRL9 swarm successor of this could do just from a psychological warfare perspective...gives me anxiety just thinking about it.
Nice. I used to work on that problem, because I was into legged running for robots and animation. Rather than putting fans on the "torso" to provide torques, I was doing that by having a leg with three joints and coordinating them to apply torques to the torso.[1] This is how far I got. (Video, generated on a Mac IIci. Hours of run time for seconds of simulation.)[2] I notice the Berkeley group is using three joints, too. They should't need those fans to exert side forces.
The real objective was to be able to handle running on slanted surfaces. That's all about slip control.
I looked at this as a two-point boundary value problem with an underactuated controller. You want to get a specific orientation, angular rate, and target position at landing. You have more target variables than you have actuators. You don't care too much about the intermediate states as long as the state at landing is what you wanted. That's rocket science. Which is good, because rocket trajectory control is a well understood problem.
Once you start solving this as a boundary value problem, you can make much more aggressive moves. Going way off balance to get a fast change of speed or direction is fine as long as the next landing is within bounds. Most running robots today start out running in place and slowly accelerate forward. They don't start by leaning forward and launching. That's because they work by trying to maintain a stability criterion. In slow robots, that's static stabiilty - CG over base. Faster ones use "ZMP", which is a generalization of static stability which adds a momentum term.
There are newer approaches, and the most recent Boston Dynamics robots are going into unstable positions to do things like flips, then recovering to a stable position. That's looking one move ahead. If you can look two moves, or footfalls, ahead, you get much more agility. Some of martial arts and dance is learning sequences of three moves ahead. That's an athletic feat.
You need two moves ahead to do broken-field running or the stop by turning that horses do, which is within the normal range of ability.
I want to encourage you to expand on what you describe as broken-field running. By two moves ahead you mean that you are planning not only for a particular path and rate but for a rate of change as well, possibly both in direction and speed.
OK. Think of it this way. Suppose you have a basic controller for a biped which can achieve a stable stand after a landing on one foot provided that the velocities are within certain limits. With a controller like that you can walk or jog a little. That's classic Asimo-level locomotion.
The limits within such a controller can recover define a target basket. If you can land within the target basket, you don't fall down. Unstable moves should end with a landing inside that target basket, leading to a stable state.
Next, you want to go from a standing start to a fast run in one stride. You lower the torso (probably by bending the knees, but it could be a piston-type motion), fall forward, and at some angle, launch. While in the air, you make the appropriate maneuvers to land on one foot within the target basket of the basic controller. Doing that is a two-point boundary value problem. That's planning one move ahead. One wildly unstable step and a controlled landing within the target basket of the basic controller. That's Boston Dynamics's flip.
Now if you could do two unstable footfalls in succession, you'd have a much more powerful system. You could do two unstable steps and then land in the target basket and let the basic controller take over for a safe stop. But you now have the option of taking one unstable step and then planning two more if possible for fast changes in speed and direction. There's always the possibility of sticking with the original plan that lands you in the target basket for a safe stop. But, while in flight for the first unstable step, you can try planning a new two-step plan. If that problem can be solved, you can continue with agile motion. If not, you fall back to the original plan and end up stopped but safe.
Two point boundary problems can be solved using known techniques. See "shooting method" and "relaxation method", both of which are optimization schemes. Solving two two-point boundary problems in sequence is harder and requires looking through a big solution space, but everybody is doing gradient descent in bumpy spaces now and it's less mysterious than it used to be.
Good PhD thesis problem for someone. Startup potential, not so much.
9 comments
[ 0.26 ms ] story [ 30.3 ms ] threadBut, then again, anybody without nukes or B-52 squadrons is out of luck.
The real objective was to be able to handle running on slanted surfaces. That's all about slip control.
I looked at this as a two-point boundary value problem with an underactuated controller. You want to get a specific orientation, angular rate, and target position at landing. You have more target variables than you have actuators. You don't care too much about the intermediate states as long as the state at landing is what you wanted. That's rocket science. Which is good, because rocket trajectory control is a well understood problem.
Once you start solving this as a boundary value problem, you can make much more aggressive moves. Going way off balance to get a fast change of speed or direction is fine as long as the next landing is within bounds. Most running robots today start out running in place and slowly accelerate forward. They don't start by leaning forward and launching. That's because they work by trying to maintain a stability criterion. In slow robots, that's static stabiilty - CG over base. Faster ones use "ZMP", which is a generalization of static stability which adds a momentum term.
There are newer approaches, and the most recent Boston Dynamics robots are going into unstable positions to do things like flips, then recovering to a stable position. That's looking one move ahead. If you can look two moves, or footfalls, ahead, you get much more agility. Some of martial arts and dance is learning sequences of three moves ahead. That's an athletic feat. You need two moves ahead to do broken-field running or the stop by turning that horses do, which is within the normal range of ability.
Fun problem, but decades later, still no market.
[1] http://www.animats.com/papers/articulated/articulated.html
[2] https://www.youtube.com/watch?v=kc5n0iTw-NU&feature=youtu.be...
The limits within such a controller can recover define a target basket. If you can land within the target basket, you don't fall down. Unstable moves should end with a landing inside that target basket, leading to a stable state.
Next, you want to go from a standing start to a fast run in one stride. You lower the torso (probably by bending the knees, but it could be a piston-type motion), fall forward, and at some angle, launch. While in the air, you make the appropriate maneuvers to land on one foot within the target basket of the basic controller. Doing that is a two-point boundary value problem. That's planning one move ahead. One wildly unstable step and a controlled landing within the target basket of the basic controller. That's Boston Dynamics's flip.
Now if you could do two unstable footfalls in succession, you'd have a much more powerful system. You could do two unstable steps and then land in the target basket and let the basic controller take over for a safe stop. But you now have the option of taking one unstable step and then planning two more if possible for fast changes in speed and direction. There's always the possibility of sticking with the original plan that lands you in the target basket for a safe stop. But, while in flight for the first unstable step, you can try planning a new two-step plan. If that problem can be solved, you can continue with agile motion. If not, you fall back to the original plan and end up stopped but safe.
Two point boundary problems can be solved using known techniques. See "shooting method" and "relaxation method", both of which are optimization schemes. Solving two two-point boundary problems in sequence is harder and requires looking through a big solution space, but everybody is doing gradient descent in bumpy spaces now and it's less mysterious than it used to be.
Good PhD thesis problem for someone. Startup potential, not so much.