Home AI News MIT Professor Develops Collision-Free Drone Fleet for Efficient Industry Operations

MIT Professor Develops Collision-Free Drone Fleet for Efficient Industry Operations

MIT Professor Develops Collision-Free Drone Fleet for Efficient Industry Operations

MIT Professor Jonathan How: Innovating Autonomous Vehicles and AI Algorithms

Jonathan How, a renowned professor at MIT, has dedicated his research to the field of autonomous vehicles, including planes, spacecraft, drones, and cars. His primary focus lies in designing distributed robust planning algorithms that enable multiple autonomous vehicles to navigate dynamic environments without colliding. How’s trajectory planning system, developed alongside his team at the Aerospace Controls Laboratory, allows a fleet of drones to operate in the same airspace while avoiding collisions. This breakthrough has significant implications for industries like agriculture and defense, promising cost savings and improved efficiency.

The Test Facility: Kresa Center for Autonomous Systems

How’s team conducts experiments at the Kresa Center for Autonomous Systems, a specialized 80-by-40-foot space with custom features for MIT’s work with autonomous vehicles. Using this facility, they test their swarm of UAVs (unpiloted aerial vehicles) to prevent collisions. To achieve this, each drone calculates its path-planning trajectory onboard and shares it wirelessly with the other vehicles. However, communication delays pose a significant challenge to multi-vehicle operations. To solve this, How’s team integrated a “perception aware” function into their system. This function allows each vehicle to use its onboard sensors to gather real-time information about other vehicles and adjust its planned trajectory accordingly. With this algorithmic fix, they achieved a 100 percent success rate in collision-free flights among their drones. The team now aims to scale up the algorithms, test in larger areas, and eventually fly outside.

From Childhood Fascination to Leading Researcher

Jonathan How’s journey to becoming a leader in his field started with his childhood fascination with airplanes. Inspired by his father’s service in the Royal Air Force, How developed a keen interest in the engineering and mechanics of flight. His academic pursuits led him to focus on applied mathematics and multi-vehicle research in aeronautical and astronautical engineering during his undergraduate studies at the University of Toronto. He continued his graduate and postdoctoral work at MIT, contributing to a NASA-funded experiment on advanced control techniques for spacecraft. Additionally, his work on distributed space telescopes during his tenure at Stanford University further shaped his career. In 2000, he returned to MIT as a faculty member.

Advancing AI Algorithms for Autonomous Vehicles

Addressing environmental factors and predicting the behavior of pedestrians are crucial challenges for autonomous vehicles, especially autonomous cars. To tackle these challenges, How’s team collects real-time data from autonomous cars equipped with pedestrian-tracking sensors. They use this data to generate models that improve the vehicles’ short-term predictions and decision-making at intersections. How acknowledges the inherent uncertainty in making predictions but emphasizes the importance of reducing that uncertainty through improved knowledge. He believes the goal is not achieving perfect predictions but gaining a better understanding of and minimizing uncertainty.

How’s research also extends to real-time decision-making for aircraft. To determine their location, identify other objects in the environment, and plan the optimal path, aircraft require quick computations, often performed 10-50 times per second. However, deploying powerful computers on small aircraft remains impractical due to cost and size constraints. How’s solution involves employing fast-to-query neural networks onboard the aircraft. These neural networks imitate the response of computationally expensive optimizers and make decisions quickly. The team trains the neural networks offline by repeatedly running an optimizer, demonstrating how to solve a task, and embedding that knowledge into the network. This approach has proven successful for UAVs, allowing them to process noisy sensory signals and quickly respond to obstacles or locate their position.

Collaboration with Industry and Solving Real-World Problems

Throughout his career, Jonathan How has collaborated closely with esteemed companies like Boeing, Lockheed Martin, Northrop Grumman, Ford, and Amazon. By partnering with industry, How focuses his research on real-world problems and leverages their expertise. His approach involves condensing industry’s complex problems into core issues, creating solutions for specific aspects, testing the algorithms in experimental facilities, and transitioning them back to the industry. This feedback loop between academia and industry fosters natural synergy and drives innovation to solve practical challenges.

Source link


Please enter your comment!
Please enter your name here