This shape-changing robot can adapt to any outdoor environment
Described as the ‘Swiss army knife’ of robots, DyRET (Dynamic Robot for Embodied Testing) is thought to be the first quadruped that can morphologically adapt its structure in-situ to efficiently traverse different 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, they’re unable to change the length of the limb, an 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 a 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 camera. An 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.
“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.
“During these outdoor experiments, DyRET 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 between. The adaptive version was significantly more effective, proving that while some changes can be catered for in the controller, robots 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.”
This methodology could be particularly crucial in an interplanetary setting says Dr Howard, with the ability to shape–shift 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).