DyRET (Dynamic Robot for Embodied Testing) starting to adapt its shape after detecting a new terrain.

Described as the ‘Swiss army knife’ of robotsDyRET (Dynamic Robot for Embodied Testing) is thought to be the first quadruped that can morphologically adapt its structure in-situ to efficiently traversdifferent outdoor environments 

This novel capability could make DyRET a crucial tool in future search and rescue, monitoring and exploratory missions, performing complex navigational tasks that ordinarily would require a team of robots working collaboratively. 

While current quadrupeds can manipulate the angle of their leg joints, theyre unable to change the length of the limban important feature needed to successfully pass over mixed media 

One of the trickiest things with robots in outdoor environments is that they need to manoeuvre over complex and unstructured surfaces, like going from a sealed road to wet grass and then to loose sand,” explains Dr David Howard from CSIRO’s Data61, who led the collaboration. . 

“To some extent, you can account for that in control element by using software to manipulate the positioning of the robot’s legs to take shorter steps on slippery surfaces, for example. But that’s still with a static body.”  

Initially created by Dr Tønnes Nygaard at the University of Oslo, Norway, and then developed and tested at CSIRO’s Data61, DyRET applies embodied artificial intelligence (AI) to interact with and learn from its surroundings.  

The robot continuously monitors the roughness and hardness of surfaces while walking using pressure sensors in its feet and a 3D depth cameraAn inbuilt machine learning model then informs the system of the most energy-efficient navigation method before adjusting the eight telescoping sections in its legs 

“The motors can change the height of DyRET by around 20%, from 60 centimetres to 73 centimetres tall,” says Dr Howard. 

morphologically adaptive robot

The morphologically adaptive robot used in this study (a) An overview of the main components of the robot. (b) The robot with the shortest (left) and longest (right) leg configuration.of the main components of the robot. (b) The robot with the shortest (left) and longest (right) leg configuration.

Just 13 centimetres could make a dramatic difference to the robot’s walk. With short legs, DyRET is stable, but slow, with a low centre of gravity. In its tallest mode, DyRET is much more unstable while it walks, but its stride length is much further, allowing it to travel more quickly. 

In outdoor comparison tests conducted by the team in 2020, this adaptive morphology method was significantly more efficient than a static body approach.  

The robot walks forward while sensing its environment, and once a terrain change has been detected, it stops walking before changing the length of its legs to the optimal morphology for the new terrain.

“During these outdoor experimentsDyRET constantly adjusted its legs as it detected the difference between terrains in response to the model calculating a lower cost of transport,” recounts Dr Howard. 

We then ran the same path using different static morphologies, setting the legs to as long and short as possible and in betweenThe adaptive version was significantly more effective, proving that while some changes can be catered for in the controllerrobots ultimately need to adapt their body to efficiently navigate a landscape.”  

“We are also looking into using more advanced methods for sensing the environment of the robot, as well as new ways to learn and adapt,” says Dr Tønnes Nygaard. “Testing it on a wider range of terrains is also something we’re working on.”

Environments and results for indoor and outdoor experiments (a) The terrain boxes used for the experiment in the controlled indoor environment. They contain sand, gravel and concrete. (b) The COT for the two best static morphologies (gravelspecialized in green and concrete-specialized in orange) and the adaptation (blue) when walking on concrete, then gravel in the boxes.

This methodology could be particularly crucial in an interplanetary setting says Dr Howard, with the ability to shapeshift in response to different surfaces, possibly a basis to mission continuity.  

DyRET’s Swiss army knife-style platform could also be well-suited to space when you consider the amount of energy and money required to lift something, which is outside of Earth’s atmosphere. If you have five robots that can perform one task collaboratively, that’s ok on Earth, but in space that’s five times the cost and weight.  

You can image these robots being equipped with welding equipment or 3D printing extrusion as they help construct a satellite, space station or base.”  


Learn more about DyRET’s proof of concept here in Nature Machine Intelligence.   

CSIRO’s Data61 would like to credit and thank the team behind DyRET; Tønnes F. Nygaard (The Norwegian Defence Research Establishment and University of Oslo), Prof Jim Torresen (RITMO, Department of Informatics, University of Oslo), Dr Charles P. Martin (School of Computing, Australian National University), Prof Kyrre Glette (RITMO, Department of Informatics, University of Oslo) and Dr David Howard (Cyber Physical Systems Program CSIRO’s Data61, Australia).