The University of Tokyo's Department of Mechanical Engineering quietly houses one of the most ambitious robotics labs on the planet. The DRAGON Lab, led by Dr. Moju Zhao, is redefining what robots can do — not by building faster drones or stronger quadrupeds, but by asking: what if the robot could change its shape mid-flight?
Three Generations of Shape-Shifting
DRAGON — The Serpent That Flies
The lab's namesake, DRAGON (Dual-rotor embedded multilink Robot with the Ability of multi-deGree-of-freedom aerial transformatiON), is an articulated chain of links, each with a pair of vectorable rotors. Think of it as a flying snake that can:
- Pass through small openings by contorting its body in 3D
- Wrap around objects to grasp them mid-air
- Reconfigure from a straight line to a U-shape to a ring — all while hovering
- Turn industrial valves while airborne — a task that took specialized ground robots until DRAGON demonstrated it
The key insight is vectorable thrust control: each link's rotor pair can tilt independently, giving the entire articulated chain full 6-DoF per link. IEEE Spectrum called it "a flying manipulator" when it first appeared.
SPIDAR — The Quadruped That Flies
If DRAGON is a snake, SPIDAR (Spherically Vectorable and Distributed Rotors Assisted Robot) is a spider. Four legs, each with a spherical rotor at the tip. It walks. It flies. It does both in the same mission.
The breakthrough: spherically vectorable thrust on each limb. Unlike traditional quadrotors where all rotors point up, SPIDAR's legs can angle any rotor in any direction — including backward and sideways during ground locomotion.
- Walk with rotors off for silent ground approach
- Fire rotors for aerial transition without dedicated flight hardware
- Use thrust vectoring to squeeze through gaps too narrow for its wingspan
- Grasp objects aerially — the 2025 ICRA paper demonstrated stable aerial manipulation
"This is not a drone with legs. This is a quadruped that learned to fly."— New Atlas, May 2025
SERPENT — Soft Bodies in the Air
The most recent direction: soft aerial robotics. SERPENT incorporates flexible passive joints that allow the airframe itself to deform during flight. Presented at ICRA 2025, this isn't just about resilience — it's about absorbing collisions, squeezing through deformable openings, and adapting to wind gusts through natural damping.
Beyond the Air: Key Research Threads
| Thread | What It Means |
|---|---|
| Hierarchical trajectory planning | Floating-base multi-link robots navigating confined spaces — 2026 IEEE TASE |
| Teleoperation with full DoF mapping | Operator controls base pose + all joint angles simultaneously — 2026 SII |
| McKibben pneumatic actuators | Deformable quadrotor that physically changes shape using air pressure — 2026 SII |
| Multimodal locomotion | Single robot walks, crawls, flies, and climbs using the same actuators |
The Hardware Stack
| Component | Approach |
|---|---|
| Actuation | Brushless DC rotors with custom gimbal mounts for vectorable thrust |
| Structure | 3D-printed carbon fiber links with integrated motor housings |
| Sensing | Onboard IMU + joint encoders + external motion capture for research |
| Control | Real-time model-predictive control (MPC) running on onboard Jetson/STM32 |
| Power | Tethered in early prototypes; moving toward onboard LiPo for untethered |
| Simulation | Custom MuJoCo + ROS 2 stacks for hardware-in-the-loop testing |
Why This Matters
The DRAGON Lab is systematically dismantling the boundary between aerial and ground robotics. Their design philosophy — "the airframe is the manipulator" — eliminates the need for dedicated arms, grippers, or separate flight/ground systems.
- Industrial inspection: A snake robot that flies into a pipe, crawls along the interior, and uses its body to seal leaks
- Disaster response: SPIDAR walks through rubble on silent legs, then takes off vertically to survey from above
- Space robotics: Articulated aerial robots for ISS interior navigation, passing through hatches of varying sizes
- Manufacturing: SERPENT-class soft aerial manipulators that reach into irregular cavities without damaging surroundings
What's Next
The lab's publication trajectory suggests three imminent directions:
- Untethered autonomy — moving from motion capture to onboard SLAM for field deployment
- Multi-robot coordination — swarms of articulated fliers that physically connect mid-air
- Learning-based control — reinforcement learning for deformable body navigation through unknown environments
Dr. Zhao's group publishes aggressively (ICRA, IROS, RAL, TASE, SII) and releases open-access preprints. The hardware designs are documented in sufficient detail for reproduction — this is one of the few places where mechanical engineering, control theory, and aerial robotics converge into machines that genuinely shouldn't exist.