Ghost Robotics’ Minitaur Demonstrates Impressive New Skills

Last time we saw Ghost Robotics’ Minitaur (which was also the first time we saw Ghost Robotics’ Minitaur), it was getting around mostly by using a sort of hopping or bounding gait. Minitaur can move fairly quickly like this, but one of the advantages that it has as a quadruped is the potential to use a variety of different gaits to help it adapt to different conditions. In a new video just posted today, Minitaur demonstrates how it’s able to handle all kinds of terrain by dynamically adjusting its gait. And it can climb. And jump. And walk on ice. And walk on two legs. And lots of other things! Ghost Robotics cofounders Gavin Kenneally and Avik De told us that one of their primary goals has been expanding Minitaur’s behaviors to allow the robot to “traverse a wide range of terrains and real-world operating scenarios,” adding that they believe “legged robots not only have superior baseline mobility over wheels and tracks in a variety of environments and terrains, but also exhibit a diverse set of behaviors that allow them to easily overcome natural obstacles.” For more details, we spoke with Kenneally and De, along with Ghost Robotics CEO Jiren Parikh. IEEE Spectrum: How does a legged robot like Minitaur compare to robots with wheels or tracks? Ghost Robotics: On flat surfaces with no objects in their path, wheeled robots are more efficient than tracked, legged, and even aerial robots. In sand, mud, and rougher terrain, tracks are superior to wheels. However, with fixed objects, obstacles, and vertical surfaces in the path of a tracked robot with no alternative path, legged robots are superior. Even if moderate objects and obstacles can be overcome by tracks, continuous unstructured terrain over a large field of operation reduces the energy efficiency of tracked devices when compared to dynamic legged robots. Another advantage is that legged robots typically have a lot more actuated degrees of freedom than similar size tracked or aerial robots, [and those additional DoF] can be recruited for tasks like reorientation, manipulation, and getting the robot unstuck in a much more flexible and versatile way. In scenarios such as sand and mudflats, tracked devices do well up to the point of getting stuck, but they have limited options for escape, whereas the Minitaur will have greater maneuverability and escape behaviors. Additionally, you would almost certainly need to attach an arm to a wheeled, tracked, or aerial robot to open a door, while we have shown the Minitaur doing this without any modifications to its body. The primary challenge in the adoption of legged robots has been the difficulty of coordinating the many degrees of freedom and balancing on a variety of terrains. Minitaur was designed specifically to allow for very flexible and versatile software control of its limbs at a high bandwidth, which gives the control designer a lot of freedom to design control algorithms that can keep improving as time goes on (without needing to modify the robot body). “The direct-drive actuators are inherently robust since there are no gears to break due to impact loading, and we have no hydraulic system or force/torque sensors that can be damaged.”What’s your experience been like with the durability of Minitaur now that you’re doing dynamic testing outdoors? Considering the Minitaur is still preproduction, we have been subjecting the robot prototype to exhaustive physical experimentation from day one, and it has been quite hard (and fun) to try and damage it in a way that isn’t easily field-repairable. Large falls can bend the aluminum legs, but those are easy to either bend back or replace without loss of functionality. The chassis, even in its current design state, has been quite robust, and with appropriate protection for the motors, the legs become the primary concern. The direct-drive actuators are inherently robust since there are no gears to break due to impact loading, and we have no hydraulic system or force/torque sensors that can be damaged. One of the core design principles of Minitaur is its reduced mechanical complexity when compared to other legged robots and tracked devices. Tracks look simple but require complex suspension mechanisms, which would be hard to repair on the fly. With regards to mobility, if a tread or suspension mechanism gets damaged on a tracked robot, it is only able to travel in circles, but if one of Minitaur’s legs are bent, or a leg is immobilized, it can continue to limp away. Image: Ghost Robotics via YouTubeCan you describe how Minitaur changes its gait to adapt reactively to different types of terrain? A very basic example is that the walk gait is designed to use feedback to react to perturbations (like the toes slipping on ice, or the uneven nature of walking on a rock bed). If you closely examine the video of Minitaur walking on ice, you can see that the legs recirculate and move much faster when they start slipping, always swinging and repositioning under the body to prevent the body from falling on the ground. A conventional way to design multilegged walking has been to use a fixed “clock” signal that moves the legs at a fixed frequency (often along a fixed trajectory). Obviously, when the legs start slipping and sliding, with a rigid locomotion architecture, it would be very challenging (if not impossible) to keep the legs under the body without feedback from the legs and environment. Your videos show Minitaur using lots of creative ways of moving across varied terrain. What kinds of multimodal locomotion are you working on? We’ve already shown fence climbing with toe attachments in our first video, and intend to demonstrate other climbing behaviors in future videos using fixed leg attachments that will support climbing various vertical surfaces. Depending on the use case, we expect to have a future design where leg attachments can be interchanged in the field. We are also working on modifications with a confidential customer to repurpose Minitaur to operate as a surface and subsurface swimmer, and as a submersible platform that would operate on a seafloor or riverbed using flipper legs. If you look at Minitaur with the legs retracted, you can see how we can make a water-sealed design with sponsons for stability and an air bladder to control buoyancy without much effort. Our robot is relatively quiet (no gearbox operating noise), which makes it useful in a variety of scientific and military applications, and it also has very high specific power (which is one of the limiting resources for underwater vehicles). “If a tread or suspension mechanism gets damaged on a tracked robot, it is only able to travel in circles, but if one of Minitaur’s legs are bent or immobilized, the robot can continue to limp away.”How far can Minitaur walk on two legs? Is there potential for it to manipulate with the other two legs while balancing? The bipedal walking is a work in progress and one of the more challenging behaviors we’re working on—we don’t think there’s currently another 3D biped in the world that uses only four actuators. However, we’ve been quite pleased with the progress we’ve made. Minitaur can take up to 20 steps [using two legs] and then drop down onto four legs when it knows it can no longer maintain its bipedal state. We’re continuing our research and intend to have it operate in a bipedal state for much longer. Using one or two of the legs for various tasks is critical behavior for the Minitaur that we are researching (for example, door opening). Bipedal use cases include object manipulation, positioning the robot for climbing a vertical surface, gaining a vantage point for a sensor reading, escape maneuvers, and bracing/flailing to aid balance. How well does Minitaur scale upward to medium-size (or larger) legged robots? We’re pushing up against fundamental limits of torque density with the commercially available electric motors we currently use. The selected motors are critical in keeping the machine at a price point that will be on par with and even below existing tracked devices, and at a much lower cost than other legged robots. With the current motor technology, we can’t build a direct-drive machine as agile as Minitaur at a length-scale much larger than Minitaur’s 40-centimeter length. We can make a heavier version at the same scale that would have better payload capability, but we’d have to make sacrifices to increase the length scale. We are also considering modified/custom motor designs in the future for specific use cases where cost is less of an issue. Having said that, we have design efforts under way that will allow us to scale Minitaur down and deliver the same functionality with a smaller chassis and payload capacity for specific use cases where a smaller form factor is necessary. Stay tuned on this front. Don’t worry, Ghost Robotics. We’ll be staying tuned. Oh, and just a friendly reminder that Minitaur is really, really relatively cheap, at just around US $10K for one, and that’s the hand-built, not-in-volume price. Ghost Robotics is working on moving Minitaur from a development platform mostly for researchers toward a commercial version, which could also include more sensors, manipulation capabilities, improved robustness, and some built-in autonomy. The fact that they’re looking to both scale the robot down a bit while actively working to decrease the cost makes me optimistic that at some point, Ghost Robotics might have a mini-Minitaur that would be accessible to hobbyists like myself. Because if I haven’t yet made it abundantly clear, I love this thing and desperately want one.