Tiny 35-Gram Drones Can Now Dodge Obstacles on Their Own
A drone lighter than a bar of soap can now think 1,000 times a second.
Deep Dive
Researchers built an FPGA-accelerated model predictive control solver, paired with a custom 6g PCB, that runs constrained MPC onboard a 35g Crazyflie drone at 1 kHz with dynamic obstacles. Their approach delivers up to 15.6x faster solve times and a 195.4x improvement in energy-delay product over state-of-the-art embedded microcontroller-based solvers, while scaling to optimization problems with over 20,000 optimization variables and a comparable number of constraints. They released the PCB design files, firmware, and FPGA solver code as open source.
Key Points
- A 35-gram drone — about the weight of a lightbulb — can now recalculate its flight path 1,000 times per second, fast enough to avoid obstacles that move.
- The new setup is 15.6 times faster and uses roughly 195 times less energy per calculation than the standard chips used in small drones today.
- The researchers published their chip design, software, and code as open source, so others can build on it — but it only works on custom hardware, not drones you already own.
Why It Matters
Longer drone flights, safer obstacle avoidance, and smarter small robots — with the designs free for anyone to build on.