Berkeley and CMU jointly develop two-footed robots with two legs walking

Remember the dynamic robots that Boston powered, avoiding obstacles, climbing stairs, or even delivering express delivery, and arbitrarily moving through various terrains.

Now, the UC Berkeley and Carnegie Mellon University labs have developed a more flexible leg-shaped robot, ATRIAS, that can walk on rough terrain like humans, across obstacles, and without Affected by the height of the obstacle and the distance between them.

That is to say, even if the ground environment is randomly changed, the foot robot can smoothly walk.

This is very powerful!

For foot-shaped robots, especially those that are currently "blind", how to "below" on the ground is strictly controlled by algorithms. These algorithms are not very friendly to the uneven ground environment.

So the biggest challenge in designing this robot is how to optimize the control algorithm for discrete footholds.

Such a flexible robot is controlled by complex dynamic equations. Specifically, the latest nonlinear control algorithm is utilized.

The latest developments made by the University of California at Berkeley and Carnegie Mellon University are a good solution to this contradiction, allowing foot robots to walk smoothly in randomly changing obstacle terrain.

Why is a foot robot?

From the Boston-powered net red robot Atlas, to Berkeley's ATRIAS, the foot robot is undoubtedly the "C-bit" in the recent robot ranks. There is a reason for this status gain.

According to the walking method, the robot can be divided into a wheel type, a foot type, a crawler type and a hybrid type. Wheeled robots are mainly suitable for flat roads, with high-speed movement performance, but can not do anything for complex terrain; crawler robots can better adapt to soft terrain, such as land, the disadvantage is that there is no power for terrain with high altitude difference; foot type The robot can adapt to almost all kinds of complex terrain. The disadvantage is that the moving speed is low and it is easy to roll over due to the center of gravity.

Most of the terrain in the world belongs to complex terrain, which has obvious advantages for complex terrain foot robots. Therefore, the research of foot robots has broad development prospects.

The most worthwhile part of the foot robot is also here – able to sail on unstructured and uneven terrain.

Figure: ATRIAS biped robots walk on random discrete terrain with different steps and steps

They are much more flexible than wheeled robots, which are difficult to navigate on terrain with large gaps or heights, and biped robots can travel through discrete and unpredictable terrain, making them space exploration and disaster response. And ideal candidates for personal robots in urban environments that need to walk on discrete terrain designed for humans, such as stairs or springboards.

In order to promote the advancement of machine motion technology, there are also many awards, such as the W award. This award is a challenge to the most advanced technology in machine motion, such as climbing stairs and stairs, driving 10 kilometers in less than 10,000 seconds, and so on.

Although the mechanical design and control strategies of foot robots have improved significantly over the years, they have not really been used in the real world. At present, the most advanced robots still have slow motion on quasi-static ground, and the anti-interference ability is weak, and the use of energy is also inefficient.

For the lower extremity exoskeleton, crossing the discrete zone is also a problem. The current solution is to add additional balancing mechanisms, such as crutches, but even this does not allow complete autonomous walking. This is also reflected in the recent Cybathlon Exoskeleton Competition.

By designing robots and feedback algorithms that enable robots to achieve accurate footprint design on complex terrain in a safe and reliable manner, we can apply this new robot to real life and turn these ideas into enhancements. Bioelectronic devices for human capabilities.

Why is dynamic walking on a springboard so difficult?

The University of Berkeley's Hybrid Robotics team has been developing a formal control framework for high-definition biped robots that not only ensures accurate step positions on discrete terrain, but also models uncertainty and external forces. Great. These methods are independent of the specific robot itself and are simulated on various robot models such as RABBIT, ATRIAS and DURUS.

In addition, these robots do not “know” what the terrain would look like in advance; only the next position will be displayed to the robot, which is a good description of what the robot might encounter in the real world.

The Berkeley team tested the control algorithm on the ATRIAS biped robot platform and was able to achieve dynamic walking on different discrete random terrains with steps ranging from 30 to 65 cm and walking height (lifting legs) of 22 cm while maintaining The average walking speed is 0.6 m/s.

It can be said that Berkeley's ATRIAS is the first time on a biped robot to complete dynamic walking on a springboard with both step and step height changes.

Why is discrete walking so difficult in robots?

First, the biped robot is a high-degree-of-freedom system whose motion is controlled by complex nonlinear differential equations that describe the hybrid dynamics of the robot's interaction with the ground: the robot must constantly contact and break contact with the surrounding environment. Interact with the environment.

In addition, ATRIAS is under-actuated, which means it has no drives at the ankle. You can imagine stepping on a springboard or stepping on a stilt to climb the stairs: the only way to maintain balance is to keep going.

The springboard problem also imposes strict restrictions on the placement of the feet. Of course, in the real world, these springboards may also collapse (we will solve this problem in the near future). In addition, the robot must also work under other physical constraints, such as motor torque limiting and friction (the robot cannot slip). All of these constraints interact to make the control design process very important.

The springboard problem has been extensively studied, with impressive results in robots like Valkyrie and ATRIAS. But the difference is that Berkeley's approach allows for dynamic walking, rather than the slow quasi-static motion that robots tend to use.

By reasoning the nonlinearities in system dynamics and taking advantage of recent advances in optimal control and nonlinear control techniques, the Berkeley team can specify control objectives in a simple and compact form while providing formal stability and safety assurance. And the desired robot behavior. This means that the robot can walk on discrete terrain without slipping or falling.

Next research

Currently, ATRIAS is still a “blind person” and needs to provide them with information about its surroundings, such as the location of the next springboard. The Berkeley team said they are now working on integrating computer vision algorithms such as depth segmentation and deep learning into controllers.

This will allow the robot to reason about the surrounding environment and develop a completely autonomous system. As a new robot named Cassie arrives at Berkeley, it plans to extend the experimental results to 3D walking on real-world springboards.

In the long run, this research will help bipedal robots navigate autonomously in indoor environments such as stairs and narrow corridors, as well as in outdoor environments such as jungle trails. Key parts of the research include safety, robustness and flexibility, which means that we want our robots to be able to walk in the "right" position to prevent them from falling, while remaining robust to unexpected external forces and disturbances. Sex.

There are many potential applications for this technology: in search and rescue, automatic humanoid robots can be deployed instead of human rescuers; when exploring unmapped/unexplored areas (such as on other planets with uneven surface heights), Or as a personal robot at home. In addition, methods developed for biped robots can also be translated into robotic devices that enhance humans, such as lower extremity exoskeletons.

Electronics Industry Follow Me

Link:Tenco