Algorithm lets drones navigate obstacles at top speed without crashing

drone speed crashing

Use of uncrewed aerial vehicles (UAV) on missions requiring both accuracy and rapidity are growing throughout business applications – and are the main focus of first responder work. Maximizing that combination of speed and precision of drones – while avoiding crashing – was also at the heart of recent promising research.

MIT algorithm helps drones navigate obstacles at maximal speed without crashing

A group of aerospace engineers at the Massachusetts Institute of Technology (MIT) studied the primary challenge UAV racing pilots face: navigating drones through a twisting course at maximum speed without the craft crashing into various obstacles. Finding a formula that ensures top velocity and safety that can be loaded on to drones, they reasoned, could help save companies operational time – and money. More importantly, it could also be the difference between life and death in emergency response scenarios. Discovering that optimal performance balance, however, proved daunting.

So when an algorithm they created to attain that idealized speed-safety objective wound up containing holes, the researchers took a hybrid route. They began pairing the myriad theoretical scenarios written into their codes with real-life experiences drone racers often must learn the hard way.

“At high speeds, there are intricate aerodynamics that are hard to simulate, so we use experiments in the real world to fill in those black holes to find, for instance, that it might be better to slow down first to be faster later,” MIT grad student and study researcher Ezra Tal told MIT News. “It’s this holistic approach we use to see how we can make a trajectory overall as fast as possible.”

Previous work has shown it’s fairly easy to develop systems that fly drones amid obstacles without crashing – as long as their speed remains relatively low. Adding velocity to the mix, however, introduces a number of factors – drag and stability, for starters – that make it harder to know just how well and quickly the craft will react. Which was why the MIT group decided to plug the gaps their entirely simulation-produced algorithm contained with experience that can only be gained behind a controller.

Enhancing tech-generated avoidance algorithms with real flight learning

To begin with, they sent computer simulated drones through a virtual obstacle course at varying speeds, and used runs without craft crashing to compose an algorithm. They then replicated the same course in concrete form, sending actual UAV programmed to fly velocities and routes from their simulations through it. Finally, they dispatched control drones through the course using standard obstacle avoidance systems, and at varying speeds, to act as de facto competitors to the experiment’s craft.

What they learned from those tests may (or not) come as a surprise to aspiring racers. UAVs trained with the MIT algorithm finished every match-up first, at times 20% faster than craft with conventional navigation software. More interesting still, the trial drones won despite defying what may be considered a logical racing decision by taking longer routes around obstacles, or considerably sacrificing speed to lower risks of crashing. An initial path chosen, for example, might have been looping compared to straight-ahead bolt by the control UAV, yet resulted in flight around hazards that improved the time-velocity-safety objective overall.

The reason for that, experts say, was algorithms in the control craft were based exclusively on simulation scenarios, while the trial drones also worked from situations from real trial flights. That meant smart decisions human pilots made while clearing obstacles at top speeds without crashing were factored in among the MIT simulations. That allowed those drones to select the best option of either computer or human sets, based on the differing situations they encountered.

“If a human pilot is slowing down or picking up speed, that could inform what our algorithm does,” Tal explains. “We can also use the trajectory of the human pilot as a starting point, and improve from that to see (things) humans don’t do that our algorithm can figure out to fly faster. Those are some future ideas we’re thinking about.”

FTC: DroneDJ is reader supported, we may earn income on affiliate links


Subscribe to DroneDJ on YouTube for exclusive videos

Show More Comments