A robotics group from Oregon, Agility Robotics, published a study detailing the latest update on its Cassie robot, a bipedal walking machine inspired by the movements of the ostrich. The equal parts awe-inspiring and creepy robot draws inevitable comparisons with Boston Dynamics' Atlas robot.
In their new paper, published in the journal Computer Science, the Agility Robotics team explained that, to their knowledge, their latest update of Cassie shows the "first controller for a bipedal, human-scale robot capable of reliably traversing a variety of real-world stairs and other stair-like disturbances using only proprioception."
Proprioception, essentially, is the ability by which we perceive our own limbs in space without using vision. So, for example, it allows us to touch our noses with our eyes closed.
So what that means is that Cassie the robot — which has already taught itself to walk using reinforcement learning — now has an incredible ability to climb stairs using only proprioceptive feedback and no pre-programmed computer model of the staircase it's climbing.
Essentially the robot is traversing the staircases blind using only proprioceptive sensors that make it aware of its immediate surroundings.
Cassie robot's proprioceptive feedback makes it a 'superhuman at stair-climbing'
Why use this method and not one that utilizes state-of-the-art liDAR sensors used in autonomous vehicles and other robots, you might ask?
This is, in part, because those systems use a great deal of computational power and are not necessarily cost-effective for a robot that is tasked with delivering boxes or parcels. Agility Robotics' traversal method, on the other hand, allows Cassie the robot to climb stairs using a fraction of the processing power otherwise required — and it does so with incredible efficiency.
As seen in the demo videos released by Agility Robotics, though the robot stumbles into railings and has to backtrack occasionally, it doesn't trip over and it invariably gets to its destination at the bottom of the stairs.
In an interview with IEE Spectrum, first author Jonah Siekmann said "we’ve joked lots of times that Cassie is superhuman at stair climbing because in the process of filming these videos we have tripped going up the stairs ourselves while we’re focusing on the robot or on holding a camera."
In order to teach Cassie how to climb stairs so efficiently, the Agility Robotics researchers used reinforcement learning to train the machine on a massive dataset of staircase scenarios, including possible obstacles encountered on the staircase.
Though Agility Robotics is a relatively new company — it was founded in 2015 — many of its researchers come from the ATRIAS project, which is one of the first to work on bipedal robots with humanlike gaits.
While there's a way to go before a robot like Cassie will be delivering groceries or assisting the elderly, efficiently overcoming one of the greatest obstacles to robot traversal — the staircase — brings us one step closer to seeing widespread consumer bipedal robots.