A robotic system developed by scientists at the University of California, Berkeley and Carnegie Mellon University's School of Computer Science enables a cheap, small-legged robot to climb and descend stairs almost its height. It can also navigate rocky, slippery, uneven, steep, and varied terrain, walk across gaps, scale rocks and curbs, and even operate in the dark.

The team put the robot through its paces by asking it to climb steps that, given its height, would be comparable to a person leaping over a hurdle, challenging it to walk through steppingstones and over slippery surfaces, and testing it on uneven staircases and hillsides in public parks. Depending on its vision and a modest onboard computer, the robot learns rapidly and masters rugged terrain.

In a simulator, the researchers used 4,000 identical copies of the robot to train it to walk and climb over difficult terrain. The robot acquired six years of experience in a single day because of the simulator's quickness. Additionally, the simulator gained motor abilities during training and stored them in a neural network that the researchers replicated in the actual robot. This method deviated from conventional ones by not requiring any manual engineering of the robot's movements.

Also Read:

India one step closer to adopting uniform charging port for electronic devices

Most robotic systems employ cameras to map their surroundings and plan their motions before carrying them out. Due to intrinsic fuzziness, mistakes, or misperceptions in the mapping step that affect future planning and motions, the process is sluggish and can frequently falter. In systems geared toward high-level control, mapping and planning are helpful. Still, they are only sometimes suitable for the dynamic demands of low-level abilities like walking or sprinting over difficult terrains.

The new approach sends vision inputs directly to the robot's control by bypassing the mapping and planning stages. The robot moves in response to what it perceives. Even the researchers are silent regarding proper leg movement. This method enables the robot to respond swiftly to approaching terrain and navigate it successfully.

The robot can be inexpensive because no mapping or planning is required, and movements are learnt using machine learning. The team's robot was at least 25 times cheaper than competing products. The team's approach may increase the accessibility of low-cost robots significantly.

This direct vision-to-control aspect is biologically inspired. Humans and animals use vision to move. Previous research from the team has shown that blind robots — robots without cameras — can conquer challenging terrain, but adding vision and relying on that vision greatly improves the system.

Also Read: Elon Musk defends $56 billion pay — 'was focused on reviving Tesla'