MIT and Penn Open-Source MIGHTY, a Drone Motion Planner That Cuts Out Six-Figure Commercial Software

Researchers at MIT and the University of Pennsylvania have released MIGHTY, an open-source motion planner that lets drones react to obstacles in milliseconds using only onboard sensors and compute. The system, detailed in a paper accepted to IEEE Robotics and Automation Letters and announced by MIT, jointly optimizes flight path and travel time in a single computational step. In simulated tests, it ran in about 90 percent of the time required by state-of-the-art methods while reaching its destination roughly 15 percent faster. In hardware tests, drones running MIGHTY hit speeds of 6.7 meters per second, or about 24 km/h (15 mph), through cluttered environments with obstacles added on the fly. The headline detail is the licensing model. Existing high-performance motion planners often depend on commercial solvers that can cost hundreds of thousands of dollars. MIGHTY is free, runs on Ubuntu 22.04 with ROS 2 Humble, and requires no proprietary dependencies.

How MIGHTY Optimizes Flight Path and Travel Time Together

Most open-source motion planners fix a travel-time estimate first, then calculate the best route within that window. MIGHTY does both at once, using a mathematical technique called a Hermite spline to produce smoother flight paths that the drone can control more precisely.

The trade-off in fixed-time planners is real. If a drone has to take a longer route around an obstacle, it must crank up its speed to meet the preset time budget. That makes the next sudden obstacle harder to dodge.

Kota Kondo, the lead author and an aeronautics and astronautics graduate student at MIT, addressed this directly in MIT’s announcement. Joint optimization gets better results, but the math problem balloons in size. The team’s fix was to skip generating each new route from scratch. Instead, MIGHTY makes an initial guess at a flight path, then refines it through iterative optimization using a lidar-built map of the scene.

The arXiv version of the paper reports a 9.3 percent reduction in computation time, a 13.1 percent reduction in travel time, and a 100 percent success rate against two state-of-the-art baselines in simulation. The code is live on the lab’s GitHub repository.

Open-Source Release Removes the Six-Figure Software Tax

MIGHTY’s release through MIT’s Aerospace Controls Laboratory eliminates the proprietary dependency problem that has limited high-performance motion planning to well-funded research groups and large commercial operators. Any developer, student, or company can now build on it.

That matters more than it sounds. The commercial solvers that high-performance planners typically depend on price out the long tail of small commercial operators and independent researchers who write the most interesting tools in this space. MIGHTY runs entirely on the drone’s onboard computer, with no cloud dependency and no remote-link reliance for the planning loop.

The research was funded in part by the U.S. Army Research Laboratory and the Defense Science and Technology Agency in Singapore. Both backers want planners that work in GPS-denied and communications-denied environments.

This is a different model than MIT’s earlier work in this lineage. The lab’s 2019 FASTER planner, also open-source, handled safety and speed using separate computations and a “stop” condition in known free space. MIGHTY rolls those into one solve.

Kondo’s Fukushima Motivation Shaped the Research

Kondo’s interest in autonomous robots tracks back to the 2011 Fukushima Daiichi nuclear accident. As a child stuck at home after the Great East Japan Earthquake, he watched workers enter radioactive areas to contain the damage and decided he wanted to build the robots that could go in instead.

That framing isn’t decoration. The applications MIT names for MIGHTY are the dangerous ones: search-and-rescue inside collapsed buildings, industrial inspection of complex structures like wind turbines, and disaster recovery in places too unstable for humans. The system is positioned for environments where a human pilot is not available and where the radio link is unreliable.

Davide Scaramuzza, who leads the Robotics and Perception Group at the University of Zurich and was not involved in the work, told MIT that Hermite splines have already proved themselves in visual SLAM, and that applying them to motion planning gives drones more freedom to compute fast, dynamically feasible motions in cluttered environments.

DroneXL’s Take

Open-source academic releases like this one matter more than they get credit for. I’ve covered MIT’s drone research for years, including the MiFly indoor RF navigation system early last year and the stormy-skies stability algorithm in June 2025. The common thread is that capability built at research universities tends to escape into the wider drone software stack within a few years. PX4, ArduPilot, and parts of the ROS 2 navigation toolchain all carry DNA from earlier academic releases. MIGHTY is the next one to watch.

6.7 m/s is fast in a cluttered space. That’s roughly the speed I’d fly a DJI Avata 2 through a wooded area while focused on not hitting branches, and a human pilot has the advantage of intent and pre-planned lines. A planner doing that autonomously while obstacles get added in real time is a meaningful result, not just a paper number.

The question MIT did not address in the announcement is integration. MIGHTY targets ROS 2 Humble on Ubuntu 22.04. Whether the major commercial drone stacks port it, fork it, or ignore it will tell us how seriously the industry takes academic open-source releases right now. The German SPRIND challenge for satellite-free navigation is pushing on a related theme from a different angle. Both lines of work matter because the future of autonomy will be solved by whoever lets the most developers contribute.

Sources: MIT News, IEEE Robotics and Automation Letters, arXiv preprint 2511.10822, mit-acl/mighty GitHub repository.

DroneXL uses automated tools to support research and source retrieval. All reporting and editorial perspectives are by Haye Kesteloo.