A robot that evolves its own hands.
We built a system where a robotic arm scans an unknown object, evolves a custom gripper through genetic algorithms, 3D-prints it, swaps it on, and picks the object up. Each generation tests 20+ designs in physics simulation, selecting the best performers to breed the next generation.
Six steps. Zero humans. Pure evolution.
From failure detection to successful grasp – the entire evolutionary cycle runs autonomously.
When rigid grippers meet complex geometry.
The robot attempts to grasp an unfamiliar object with its current gripper – and fails. Traditional systems would stop here, requiring human intervention. Our system recognizes the failure and automatically triggers the evolutionary pipeline to design a custom solution.
It sees geometry the way an engineer would.
The depth camera captures point clouds and reconstructs watertight meshes. PCA analysis identifies principal axes – the system automatically reorients objects so the shortest dimension becomes the grip axis. Extracts volume, extents, and surface normals to inform gripper generation.
Natural selection for robot hands.
A 6-parameter genome encodes each gripper design. Population of 20 individuals per generation. Top 20% survive (elitism), rest breed through tournament selection, crossover, and mutation. Physics simulation is the fitness judge – only designs that can lift the object survive to reproduce.
Direct mesh-to-machine loop.
The winning STL from evolution is automatically processed by Bambu Studio. Support structures, orientation, and slicing parameters are selected based on the gripper’s mechanical requirements. PLA prints in 2-4 hours depending on complexity.
Hot-swapping evolved hardware.
The robot autonomously removes the old gripper and attaches the newly printed, evolved design. Our custom tool changer enables rapid finger swaps without manual intervention. The system verifies proper attachment before proceeding to the grasp attempt.
The loop closes – autonomously.
With the evolved gripper installed, the robot re-attempts the grasp. The evolutionary cycle completes when the robot successfully performs the task it previously failed. If it still fails, the system can restart evolution with adjusted constraints or continue for more generations.
Where adaptive gripping matters.
Manufacturing
Mixed-product lines where changeover costs kill margins. One robot handles infinite part geometries.
CNC & Machining
Automated workpiece loading for complex geometries. No jig redesign between runs.
Research & Education
A platform for studying embodied intelligence, morphological computation, and adaptive robotics.
E-commerce Logistics
Adaptive bin-picking for warehouses handling thousands of unique SKUs daily.
Let’s build the next generation of adaptive robotics.
Seeking industry partners for deployment and collaboration.