The Atlas robot of the famous Boston Dynamics robotics company turns ten years old this year. And during those ten years, we've watched as this 1.5-meter, 89-kilogram robot android, powered by 28 hydraulic actuators, has gained new capabilities and improved its movements. And now Boston Dynamics engineers have begun to teach the robot how to interact with its environment and use improvised objects to perform its task.
In a recently published video the Atlas robot has to deliver a bag of tools to a live "colleague" on a makeshift "construction site". To perform this action, the robot needs to take a board and use it as a "bridge". The robot then has to lift the heavy bag of tools, climb one level up, and toss the bag another level up.
After completing the main task, Atlas does something that would get any worker on a real construction site fired. He throws down a wooden box, jumps on it and makes his "crown" jump with a flip and a 540 degree turn. Note that according to the information available, the robot "uses the full power of every joint and actuator" to make this jump. In other words, this jump is the technical limit of the robot's capabilities.
Despite the relative simplicity of all the actions, behind their execution there is a huge amount of painstaking work done by Boston Dynamics specialists, and the second video sheds light on it all. It turns out that there is no need to program each individual movement of each part of the robot body, as it was before. Now the robot is only given generalized tasks - picking up a board and using it as a bridge, lifting a tool bag, going up one level, dropping the bag, knocking over a box and doing a rollover jump.
Everything else, including determining the location, shape and orientation of objects in the environment, the robot does by itself, using cameras, a laser scanner and other sensors.
Moreover, a lot of effort by Boston Dynamics specialists was spent on equipping the robot to do things that people do automatically at a subconscious level. Consider, for example, the very simple task of picking up a bag of tools for a human. Atlas finds a bag lying on the floor, and as it approaches it, it builds a model of further actions. In doing so, he finds answers to the following questions on his own as he performs his actions: How much does the bag weigh? Where is its center of gravity? Where is the best place to grab the bag? At what angle should the hands and grips be positioned? How do I prepare my body to lift the bag? How do you hold the bag during the move and before the throw?
far from perfect and from being usable as a practical employee of any kind. And only Hyundai's large financial resources allow them to continue now their painstaking, innovative and very interesting work, the results of which will surely be used in the next generations of robotics.