Researchers at MIT have developed an autonomous robot that can effectively navigate through pedestrian traffic.
An autonomous robot has to solve these four challenges in order to effectively navigate through a pedestrian traffic:
It has to know its precise location in this world.
It has to have the capacity to recognize its environment.
3. Motion Planning
It has to be able to identify the best path towards a given destination.
It must be able to control its body to physically execute the desired path.
To solve the problem of localization, the MIT researchers equipped their robot with open-source algorithms to map its environment and verify its position. To solve the problem of perception, the researchers equipped the robot with webcams, high-resolution lidar sensor and a depth sensor. To solve the issue of control, the researchers took advantage of the standard methods used to drive autonomous vehicles.
In solving the motion planning issue, the researchers needed to innovate. Pedestrians rarely walk in a straight pattern. People sometimes veer off their normal course and change directions. There are also norms that the robot has to abide by such as maintaining a respectable berth, keeping to the right and passing on the left.
To solve the motion planning issue, the researchers trained the robot using reinforcement learning. They trained it to take certain paths, taking into consideration the speed and trajectory of other objects in the environment. The researchers also trained the robot in simulations to follow pedestrian rules. For instance, the researchers reward the robot if it passes people on the left, and penalizes it if passes people on the right side.
Yu Fan “Steven” Chen, the lead developer of the robot, told MIT News that a “socially aware” robot is useful for package and food delivery, and for personal mobility devices to transport people in large crowded spaces such as hospitals, airports and shopping malls.