.jpg)
A novel technique has been created to securely drive an autonomous robot without knowing its surroundings or the size, form, or placement of obstacles it may meet.
An autonomous spacecraft exploring the cosmos descends through the atmosphere of a distant exoplanet. The robotic vehicle, as well as the researchers that programmed it, are unfamiliar with the terrain.
How can the spaceship plan a safe route with so much uncertainty? How can it avoid being smashed by a randomly shifting barrier or blown off track by sudden gale-force winds?
Researchers at MIT have developed a novel technology that could aid in the safe landing of this spacecraft. Their method could allow an autonomous vehicle to calculate a provably safe course in extremely uncertain settings involving various uncertainty in both environmental variables and potential collisions with objects.
The method could potentially aid a vehicle in navigating around obstacles that move in unpredictable ways and change shape over time. Even though the vehicle's starting place is unknown and it is unclear how the vehicle will proceed owing to environmental disturbances like wind, ocean currents, or rugged terrain, it calculates a safe trajectory to a designated region.
According to co-lead author Weiqiao Han, a graduate student in the Department of Electrical Engineering and Computer Science and the Computer Science and Artificial Intelligence Laboratory, "this is the first technique to address the problem of trajectory planning with many simultaneous uncertainties and complex safety constraints" (CSAIL).
“Future robotic space missions need risk-aware autonomy to explore remote and extreme worlds for which only highly uncertain prior knowledge exists. In order to achieve this, trajectory-planning algorithms need to reason about uncertainties and deal with complex uncertain models and safety constraints,” says co-lead author Ashkan Jasour, a former CSAIL research scientist who currently works on robotics systems at NASA's Jet Propulsion Laboratory (JPL).
Brian Williams, a professor of aeronautics and astronautics and a member of CSAIL, is a co-author on the study with Han and Jasour. The research has been nominated for an exceptional paper award and will be presented at the IEEE International Conference on Robotics and Automation.
Avoiding assumptions
Other approaches for selecting a safe path forward include assumptions about the vehicle, obstacles, and environment because this trajectory planning problem is so complex. According to Jasour, these methods are too rudimentary to be used in most real-world situations, and as a result, they cannot ensure that their trajectories are safe in the presence of complicated and unknown safety restrictions.
“This uncertainty might come from the randomness of nature or even from the inaccuracy in the perception system of the autonomous vehicle,” Han says.
Instead of estimating the exact environmental conditions and placements of barriers, they built an algorithm that calculates the likelihood of seeing different environmental circumstances and impediments at different locations. These calculations would be done with the help of a map or photographs of the surroundings provided by the robot's vision system.
Their algorithms treat trajectory planning as a probabilistic optimization issue using this methodology. This is a mathematical programming framework that allows the robot to fulfill planning objectives like maximizing velocity or minimizing fuel consumption while also taking into account safety limitations like avoiding obstacles. According to Jasour, the probabilistic algorithms they built reason about risk, which is defined as the likelihood of failing to meet the safety limits and planning objectives.
This probabilistic optimization is too complicated to perform using typical approaches since the problem incorporates multiple uncertain models and constraints, ranging from the location and shape of each barrier to the robot's starting location and behavior. The researchers converted the probabilistic optimization into a more straightforward, simpler deterministic optimization problem that can be handled effectively with existing off-the-shelf solvers using higher-order statistics of probability distributions of the uncertainty.
“Our challenge was how to reduce the size of the optimization and consider more practical constraints to make it work. Going from good theory to good application took a lot of effort,” Jasour explains.
The optimization solution creates a risk-bounded route, which means that if the robot follows it, the chances of colliding with any barrier are less than a given threshold, such as 1%. They can then generate a set of control inputs that will safely guide the car to its destination.
Charting courses
Several simulated navigation scenarios were used to test the approach. They modeled an underwater vehicle charting a journey from an unknown location to a goal zone, passing through a series of oddly shaped obstacles. It was able to hit the target safely 99 percent of the time. They also utilized it to create a safe route for an aerial vehicle that avoided many 3D flying objects with unknown sizes and positions that could move over time, all while dealing with severe gusts. The airplane reached its target region with a high likelihood using their technology.
The algorithms required anywhere from a few seconds to a few minutes to build a safe trajectory, depending on the complexity of the surroundings.
According to Jasour, the researchers are now working on more efficient techniques that would considerably cut runtime, allowing them to move closer to real-time planning scenarios.
Han is also working on feedback controllers for the system, which will assist the vehicle stay closer to its planned trajectory even if it deviates from it at times. He's also working on a hardware implementation that will allow the researchers to show off their technique on a real robot.