Robot Design Using Computer-aided Creativity

If you need a robot that climbs stairs, what shape do you think it should have? Should it be shaped like a human with two legs, or like a spider with 8 legs?

Choosing the right shape for a robot is crucial for the robot to be able to move across a specific terrain. It is impossible to build and carry out experiments for every potential form of the robot. But now, a system developed by MIT has made it possible to simulate and see which design is the most effective.

You begin by informing the system, called RoboGrammar, the kind of robot parts you have around you. You also need to tell it what kind of terrain the robot should be able to traverse. RoboGrammar finishes the job, generating an optimized structure and control program for the robot.

This innovation could introduce computer-aided creativity into the field. ‘Robot design is still a very manual process’, Allan Zhao says. Zhao is the lead author of the paper and a PhD student in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). According to him, RoboGrammar is a ‘way to come up with new, more inventive robot designs that could potentially be more effective’.

The paper will be presented at the next SIGGRAPH Asia conference. Some of the coauthors include, PhD student Jie Xu, Postdoc Mina Konakovic-Lukovic, Postdoc Josephine Hughes. PhD student Andrew Spielberg and Professors Daniela Rus and Wojciech Matusik.

Robots are built with the capacity for an almost infinite variety of tasks, yet ‘they all tend to be very similar in their overall shape and design’, Zhao says. For instance, ‘when you think of building a robot that needs to cross various terrains, you immediately jump to a quadruped (a four-legged animal). We were wondering if that’s really the optimal design’.

Zhao and his team presumed that a more innovative design could advance its efficiency. So they built a computer model for this job, one that wasn’t unduly influenced by convention. While the goal was to be as inventive and innovative as possible, Zhao had to set some ground rules.

The universe of possible robot forms is ‘primarily composed of nonsensical designs. If you can just connect the parts in arbitrary ways, you end up with a jumble’, Zhao says in his paper. To avoid that, his team developed a ‘graph grammar’— a set of restrictions on the order of a robot’s component. These rules will make sure that all the computer-generated designs work, at least at the most basic level.

Zhao says the rules of his graph grammar were inspired by animals— specifically the arthropods. These include spiders, insects and lobsters. Arthropods account for more than 80% of the known animal species. ‘They’re characterized by having a central body with a variable number of segments. Some segments may have legs attached’. Zhao says. ‘And we noticed that that’s enough to describe not only arthropods, but more familiar forms as well’.

Zhao adopted the arthropod-inspired rules, though he made some mechanical adjustments. For instance, the computer conjured wheels instead of legs.

Making use of Zhao’s graph grammar, RoboGrammar operates in a three-step process:

• Defining the problem.

• Drawing up possible robotic solutions.

• Selecting the optimal ones.

The problem definition is usually done by the human, who inputs the available robotic components. ‘That’s key to making sure the final robots can actually be built in the real world,’ Zhao says. The human also specifies the kind of terrain to be navigated and may include combination of elements like steps or flat areas.

With these inputs, RoboGrammar makes use of the graph grammar to design thousands of possible robot structures. They look like racecars, spiders or even a person doing a push-up. ‘It was pretty inspiring for us to see the variety of designs,’ Zhao says. ‘It definitely shows the expressiveness of the grammar’. Though the grammar cranks out a lot of designs, they aren’t always of good quality.

Choosing the most effective robot design requires controlling the movement of each robot and analyzing its function. ‘Up until now, these robots are just structures’, Zhao says. The controller are the instructions that gives those structures life, and govern how the robot’s various motors move. Zhao’s team developed a controller for each robot with an algorithm called Model Predictive Control, whose priority is swift forward movement.

‘The shape and the controller of the robot are deeply intertwined,’ Zhao says, ‘which is why we have to optimize a controller for every given robot individually.’ Once the simulated robots can move about freely, the researchers look for high-performing robots with a ‘graph heuristic search’. This neural network algorithm samples and analyzes the robots and learns what designs will be more suited to specific tasks. ‘The heuristic function improves over time’, Zhao says, ‘and the search converges to the optimal robot.

This all occurs before the human designer even picks up a screw.

‘This work is a crowning achievement in the 25-year quest to automatically design the morphology and control of robots’, Hod Lipson who was not a participant in the project says. ‘The idea of using shape-grammars has been around for a while, but nowhere has this idea been executed beautifully as in this work. Once we get machines to design, make and program robots automatically, all bets are off.’

It is Zhao’s intention that the system be a spark for human creativity. He describes RoboGrammar as a ‘tool for robot designers to expand the space of robot structures they draw upon’. Zhao also included that the system could be adapted to attain other goals than just navigating terrains. He also says that RoboGrammar could help populate virtual worlds. ‘Let’s say in a video game you wanted to generate lots of kinds of robots, without an artist having to create each one’, Zhao says. ‘RoboGrammar would work for that almost immediately’.

A surprising result of the project? ‘Most designs did end up being four-legged in th end’, Zhao says. ‘Maybe there really is something to it’.

By Marvellous Iwendi.

Source: MIT News