A vast number of applications for legged robots entail tasks in complex, dynamic environments. But these environments put legged robots at high risk for limb damage. This paper presents an empirical study of fault tolerant dynamic gaits designed for a quadrupedal robot suffering from a single, known "missing" limb. Preliminary data suggests that the featured gait controller successfully anchors a previously developed planar monopedal hopping template in the three-legged spatial machine. This compositional approach offers a useful and generalizable guide to the development of a wider range of tripedal recovery gaits for damaged quadrupedal machines.

Artificially intelligent (AI) systems have imbued robots with the ability to grasp and manipulate objects with humanlike dexterity, and now, researchers say they've developed an algorithm through which machines might learn to walk on their own. In a preprint paper published on Arxiv.org "Deep reinforcement learning can be used to automate the acquisition of controllers for a range of robotic tasks, enabling end-to-end learning of policies that map sensory inputs to low-level actions," the paper's authors explain. "If we can learn locomotion gaits from scratch directly in the real world, we can in principle acquire controllers that are ideally adapted to each robot and even to individual terrains, potentially achieving better agility, energy efficiency, and robustness." The design challenge was twofold.

Designing agile locomotion for quadruped robots often requires extensive expertise and tedious manual tuning. In this paper, we present a system to automate this process by leveraging deep reinforcement learning techniques. Our system can learn quadruped locomotion from scratch using simple reward signals. In addition, users can provide an open loop reference to guide the learning process when more control over the learned gait is needed. The control policies are learned in a physics simulator and then deployed on real robots. In robotics, policies trained in simulation often do not transfer to the real world. We narrow this reality gap by improving the physics simulator and learning robust policies. We improve the simulation using system identification, developing an accurate actuator model and simulating latency. We learn robust controllers by randomizing the physical environments, adding perturbations and designing a compact observation space. We evaluate our system on two agile locomotion gaits: trotting and galloping. After learning in simulation, a quadruped robot can successfully perform both gaits in the real world.

Shortly after SoftBank acquired his company last October, Marc Raibert of Boston Dynamics confessed, "I happen to believe that robotics will be bigger than the Internet." Many sociologists regard the Internet as the single biggest societal invention since the dawn of the printing press in 1440. To fully understand Raibert's point of view, one needs to analyze his zoo of robots which are best know for their awe-striking gait, balance and agility. The newest creation to walk out of Boston Dynamic's lab is SpotMini, the latest evolution of mechanical canines. Big Dog, Spot's unnerving ancestor, first came to public view in 2009 and has racked up quite a YouTube following with more than six and one half million views.

Ghost Robotics--a leader in fast and lightweight direct-drive legged robots--announced recently that its Minitaur model has been updated with advanced reactive behaviors for navigating grass, rock, sand, snow and ice fields, urban objects and debris, and vertical terrain. The latest gaits adapt reactively to unstructured environments to maintain balance, ascend steep inclines up to 35º, climb up to 15cm curb-sized steps, crouch to fit under crawl spaces as low as 27cm, and operate at variable speeds and turning rates. Minitaur's high-force capabilities enable it to leap up to 40cm onto ledges and across gaps of up to 80cm. Its high control bandwidth allows it to actively balance on two legs, and its high speed operation allows its legs to navigate challenging environments rapidly, whilst reacting to unexpected contact. "Our primary focus since releasing the Minitaur late last year has been expanding its behaviors to traverse a wide range of terrains and real-world operating scenarios," said Gavin Kenneally, and Avik De, Co-founders of Ghost Robotics.

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 walk on two legs. 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."

Minitaur is designed to be an affordable and practical quadruped robot. Not only can it cross obstacle-strewn terrain impassable to wheeled and tracked robots, it can also climb stairs and chain-link fences. Jiren Parikh, CEO of its manufacturer, Ghost Robotics, says it can even clamber up trees. The current version of Minitaur weighs 6 kilograms and can crawl, sidle crabwise or rear up against a vertical surface just like an excitable dog to reach high objects. Attachments on the end of the robot's legs allow it to open door handles as well as to grip and climb fences.

A new robotics startup has launched a sprightly little quadruped robot that is able to scale fences, manipulate door handles, turn back flips, climb stairs and jump up and down while keeping its balance, almost like a robot version of Jiminy Cricket. The Minitaur is a four-legged robot measuring 15.6in by 10.8in that weighs just over 5kg and can carry a payload of over 3kg. The robot features patent-pending gearless direct drive motors that behave like springs and a specialised leg design with sensors that work together to provide precise force feedback, so that the robot can balance and reorient itself from a fall while running and jumping over difficult terrain. Powered by a 72MHz Arduino-compatible robot microcontroller, the robot features high speed and high resolution encoders that enable the robot to sense the ground and process the feedback from the sensors and adapt to situations in real time to keep itself from over-balancing. Minitaur currently retails for 10,000 ( 7,670) and has a maximum running speed of 2m/s and a turning speed of 1 rad/s.

Bounding around like an excitable puppy, this amazing four legged robot can open doors and even climb over fences by itself. The Ghost Minitaur can be programmed with a series of different gaits from bounding, trotting, and walking to something called pronking – the kind of four-legged spring seen in gazelles. The miniature robot, which measures just 1.3 feet long (40cm) – about the same as a small dog – has a top speed when running of just 4.5 mph. The team behind the Minitaur are currently selling them for around 10,000 a go but they hope to reduce the price as demand increases. The developers have used a series of direct drive electric motors and some custom electronics to allow the robot to move in a range of different ways.

Bipedal and quadrupedal locomotion has been an ongoing challenge for robots. There's been a lot of progress over the last few years, though, especially when it comes to dynamic motions: not just walking without falling over, but also climbing, running, jumping, and more. This is where the real value of legs is: they enable robots to deal with the kinds of obstacles and terrain and situations that wheels and tracks can't. Getting quadrupeds to do these kinds of useful and fun things requires that a.) you know what you're doing and b.) you have a robot that can do what you want it to do. Unfortunately, building legged quadrupeds is difficult, expensive, and time consuming.