For robots, it is a time-consuming and power-hungry operation to choose a collision-free path. Roboticists are considering various algorithms along with a combination of sensors to make their robots capable of avoiding obstacles in real time.
In general, finding a collision-free path, called motion planning, is a real challenge for roboticists. This is even more challenging when you have a tight budget for power consumption and computational resources.
However, sometimes it's possible to apply some limitations to a general problem and find practical solutions. For example, researchers from Duke University employed pre-computation along with parallelism to expedite real-time motion planning by a factor of 10000.
Example of Duke's collision-avoidance planning to choose a safe path. Image courtesy of Duke University (PDF).
This was achieved at the cost of the necessity of examining the environment before the robot started moving.
The Aggressively Flying Quadrotor
Recently, Vijay Kumar’s lab at the University of Pennsylvania in cooperation with researchers from Qualcomm has unveiled a quadrotor which can fly aggressively through a window. You may think that you have seen similar robots before; however, there is a big difference between previously designed robots and this new technology.
Generally, to exhibit challenging maneuvers, a quadrotor depends on an array of cameras mounted on the walls and some external processors. The image captured by the cameras is processed and the outcome is delivered to the robot. The computer can issue precise commands and the only thing that the robot needs to do is to follow the orders. However, the new robot performs both the image capturing and processing onboard.
The quadrotor carries an IMU, a Qualcomm Snapdragon, and Hexagon DSP. With the onboard sensors and processors, the robot is able to perform localization, state estimation, and path planning autonomously.
The research team notes that the robot, which is the result of a six-year endeavor, has miniaturized whole the camera array and external processors to a 250-gram quadrotor. They hope that this technology could bring the quadrotors out of the carefully controlled labs and into the environments where they can be really helpful.
A paper describing the aggressive quadrotor, “Estimation, Control, and Planning for Aggressive Flight With a Small Quadrotor With a Single Camera and IMU”, is submitted to Robotics and Automation Letters and ICRA 2017.
Another interesting robot designed by Kumar’s lab is a bio-inspired robot which addresses the problem of motion planning from a more radical point of view. This technology shows how considering the application can lead to a more efficient design.
A Robot that Bounces Back
A group of researchers at the University of Pennsylvania’s GRASP Lab has decided to design robots which can tolerate collisions without being damaged. Since the robot simply bounces off the obstacle, they do not have to find a collision-free path. They are following the idea that “it’ll be fine” even if a robot collides with an object. This “it’ll-be-fine” philosophy can dramatically simplify the utilized algorithms. The robot may crash softly into objects because it is designed in a way that there will be no damage. After a few trials, the robot will be able to reach its target location.
The idea is that this is an imitation of how small flying insects like bees find their way around some objects on their path. In other words, since the sensors and controllers utilized in small robots are not precise enough to avoid collisions, the researchers have focused on a design which does not get ruined as it hits an object.
These robots hope to emulate how bees fly and bounce back after collisions.
The experimental bio-inspired quadrotors of the UPenn group are 25-gram, 10-centimeter-wide pico quads. The robots have a self-righting roll cage which is made from a yarn consisting of 12000 strands of carbon fiber.
The robot’s controller is very simple and does not consider the location of the other pico quads or obstacles. The only goal of the controller is to develop the ability to recover from collisions and provide a stabilized flight which eventually direct the robot towards a target location. In this way, researchers managed to avoid solving the challenging problem of motion planning.
Kumar calls the new method quite radical because it only needs the local information of the robot and not the location of the obstacles. This method can lead to smart robots which are able to navigate cluttered indoor environments––a feature which is very helpful in a search-and-rescue mission. In such missions, a swarm of the flying robots could enter a target building and provide a map of the inside.
A paper describing the new technique, “Bio-inspired Swarms of Small Aerial Robots”, is submitted to Interface Focus.
Featured image is a screengrab used courtesy of Vijay Kumar.