M4 (Multi-Modal Mobility Morphobot) specs

Caltech CAST and Northeastern University (USA) M4 (Multi-Modal Mobility Morphobot) (1)

Caltech CAST and Northeastern University (USA) M4 (Multi-Modal Mobility Morphobot) (2)

Caltech CAST and Northeastern University (USA) M4 (Multi-Modal Mobility Morphobot) (3)

Not for saleUniversity research prototype, first described in a 2023 Nature Communications paper.

Not for sale (research)Not programmableWheels become rotors8 modesSelf-righting

Not for saleShipping: Not for sale. A research prototype that was never sold.

Remote control: Not possible. Research prototype.

Nature Communications paper (2023) ↗Caltech news, M4 ↗Caltech news, humanoid and M4 demo ↗ATMO paper, Communications Engineering (2025) ↗

Overview

M4 is a four-wheeled research robot whose propeller shrouds are also its wheels, so it can drive, then swing its legs up and fly as a quadcopter. It can also crawl, balance upright on two wheels to look around, tumble over large obstacles and climb steep slopes with thruster help. A follow-up, ATMO (Caltech, 2025), changes from drone to car shape in mid-air before landing. In a 2025 Caltech and TII demo, a Unitree G1 humanoid carried M4 on its back and launched it.

Specifications

Category	Drones · Fly + drive or walk
Released	2023
Country	USA
Availability	Not for sale (research)
Programmability	No programmability
Built-in autonomy	Navigates · Goes to places on its own, such as mapped rooms, waypoints or its charging dock
LLM support	None · No built-in LLM and no SDK to connect one.
Best for	Professionals & labs
Vision sensors	Depth camera
Actuators	Mixed
Audio	No audio
Compute type	NVIDIA Jetson, Arduino / ESP32 / microcontroller
Remote control	Not programmable
Size	About 0.7 m long, 0.35 m wide and tall in driving mode (1.0 m tall when balancing on two wheels), about 6 kg
Drive	Four shrouded propellers whose shrouds double as 0.25 m wheels, each with its own geared wheel motor. Eight hip servos swing the legs to switch between driving, flying, walking, balancing and tumbling.
Compute	Nvidia Jetson Nano for path planning, plus an STM32H7 flight controller and a SAM3X8E main controller
Vision	Intel RealSense D455 stereo depth camera
Lidar	None
Sensors	IMU, motor encoders, current and voltage sensors
Runtime	Not published
Speed	Not published. Climbs slopes up to 45 degrees with thruster help.
SDK	None public. Simulation and path-planning code available from the authors on request.