Tongji drone swarm maps park woods without ground control; team withholds code
A trio of quadcopters from Tongji University in Shanghai surveyed a stand of trees in a park without a pilot or any ground station, according to a peer-reviewed paper in the Springer journal Autonomous Intelligent Systems. The authors have not released the code behind the system, DroneXL reported.
The framework, called ESCAPE, was described in a paper published on December 22, 2025, by authors at Tongji, Tianjin University and the Beijing Institute of Technology. Each drone was fitted with an Intel NUC computer and a Livox Mid-360 lidar, and all routing and teammate-avoidance decisions were made on board, DroneXL reported.
How the swarm divides the work
The swarm splits its search area into grid cells, and the drones trade cells in pairwise exchanges with no central hub giving orders. Each drone then chooses a visiting order for its cells and builds a 3D map as it flies at about 1.0 meter per second.
ESCAPE builds on Racer, a framework from the Hong Kong University of Science and Technology whose code is on GitHub. The main change to the routing step is a penalty meant to stop a drone from hovering and shuttling in front of a large obstacle that spans several cells, a behavior the authors say Racer shows.
For collision avoidance, each drone broadcasts its position, and nearby drones compute a risk score from distance, heading and closing speed. Above a threshold, the lower drone hovers while the higher one replans a route around it, the paper says.
Tests and limits
In the outdoor test, three drones covered a 12-by-20-meter patch of trees up to a height of 2 meters. Indoors, four drones mapped a 12-by-30-meter exhibition hall containing columns and a display board.
Four quadcopters, circled in red, fly between concrete columns and a display board in the glass-walled exhibition hall. Photo: DroneXL
In simulation across three office maps with two, four and six drones, ESCAPE finished exploring sooner than Racer on average in seven of the nine setups, and sooner than the Role framework from ETH Zurich in every one, according to the paper. Both simulation studies assumed perfect localization and communication, DroneXL noted.
The physical flights ran over a real network, and the team says they avoided collisions without noticeable network delays. According to DroneXL, the paper offers no collision tallies, packet-loss data or success rates for those flights.
The paper's code availability statement says the code "is not available for public release" because of "the confidentiality requirements of the research." It does not say whose requirement that is. The work was funded by Chinese public sources, and the paper describes only civilian uses such as environmental monitoring and forest search and rescue, DroneXL reported.
