Climate
Privacy
Bob Dylan
Drones
Data Centers
AI Governance
2026 Maine Senate Race
Quadrotors Master Narrow-Gap Flight
Coverage from Science, Embodied Global, and others
Articles
3
Active Days
14
The Topic
Zhejiang University researchers developed a sensorimotor control policy that converts onboard camera and flight-state data directly into motor commands, allowing quadrotors to traverse narrow, tilted, and moving gaps without GPS, motion capture, or prior knowledge of gap position. Trained with reinforcement learning in simulation and supported by model-based trajectory initialization, the system was demonstrated in real-world tests, including a reported 5-centimeter-clearance passage and maneuvers at tilt angles up to 90 degrees. The work could improve drone operations in confined environments such as collapsed structures, industrial sites, tunnels, and pipelines, while its broader operational reliability remains to be established.
First Article: 06/10/26
Latest Article: 06/23/26
Summary
- Quadrotors used onboard vision and flight-state measurements to generate low-level motor commands directly.
- The system reportedly traversed a rectangular gap with 5-centimeter clearance and handled gap tilts of up to 90 degrees.
- Demonstrations included gaps of different shapes, closely spaced tracks, and moving openings.
- The policy operated without GPS, motion capture, or prior knowledge of a gap’s location and orientation.
- Training combined reinforcement learning and model-based trajectory initialization to address difficult constrained-flight maneuvers.
- Real-world sim-to-real validation suggests potential for rubble search, industrial inspection, pipeline work, and confined-space exploration.
History
The update mainly adds publication and validation detail: the work is now explicitly tied to Science Robotics and presented as tested in real-world conditions. It also reframes the system as direct sensor-to-motor control without prior gap-location knowledge.
The story is largely stable, but the framing has shifted slightly toward a more application-oriented account of the drone system's practical value in confined and damaged environments. The current version also more explicitly states the sensing stack and confirms the work's validation without changing the core technical claim.
Featured
Timeline: 14 Days
Hover over any logo to see coverage summary, click for full article.