I Built an Even Better Ropebot Dog
Ayyub Musa, an engineering student and robotics creator, describes building a lower-cost, lighter capstan-driven robot dog with his senior design team.
In this talk
- 0:00
Turning Cara into a team project
Musa traces his robot dog projects back to a childhood wish for a German Shepherd. His earlier dog, Cara, used capstan drives, which rely on tensioned rope around smooth drums to create responsive joints. For his senior design project, he teamed up with Jack Hayman, Pranava, Megan, and Derek Gabby to improve Cara’s cost, weight, and construction.
- 2:36
Designing an affordable actuator
A customer survey pointed to three priorities: a price near $1,000, weight below 20 pounds, and durability. Since Cara cost about $3,000, the team targeted a major price reduction by developing inexpensive, high-performance actuators. They rewound drone motors by hand, tripling their torque, and built a capstan actuator rated for 12 newton-meters of peak torque at a cost of $80.
- 7:34
Reworking the leg and controls
The revised leg keeps a coaxial five-bar layout while simplifying assembly, reducing fastener variety, and replacing the fragile TPU foot with a squash ball. The team wrote inverse-kinematics equations to move the foot to set positions and generate cycloidal steps. Because the joints lack absolute encoders, each leg first finds its position by moving to its limits and detecting increased motor current.
- 13:38
Finding and fixing a gait problem
The team aimed to improve walking with shorter, quicker steps, better balance calculations, and faster turns. When the robot consistently veered left, software adjustments did not solve it. They traced the cause to identical legs creating mechanical asymmetries across the robot, then rebuilt the legs in symmetrical versions.
- 19:22
Results and next steps
Cara 2.0 came in more than 50% cheaper than the earlier design and weighed about 13 pounds less. It can walk at a nominal 1.8 feet per second, jump 4.5 inches, and carry 15 pounds. Musa says future work could add reinforcement learning and camera or lidar perception; he also made build instructions, CAD, and code available through Patreon and his website.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.