Autonomous Food Prep Robot

As part of Cheforge, I led the mechanical design of our autonomous food-prep robot, a system built to automate the entire prep workflow in restaurants. The robot’s purpose is to wash, peel, cut, and portion ingredients automatically, reducing labor cost, eliminating repetitive prep work, and increasing consistency.

My role centered on mechanical R&D, covering mechanism design, prototyping, and iterative testing to drive the robot from concept to dependable hardware. This included defining the industrial aesthetic, engineering the robot’s structural foundation, and developing the mechanisms that perform the core prep tasks. I also supported the VLA training pipeline by teleoperating dual-arm demonstrations used to fine-tune manipulation behaviors.

Note: Many core mechanisms and system-level components are not shown due to ongoing proprietary development at Cheforge.

🟩 Mechanical R&D:

Injection-Molded Body & DFM/DFA

Designed the exterior shell with uniform wall thickness, proper draft angles, and simplified geometry to meet injection-molding constraints and reduce tooling complexity.
Integrated mounting features, ribs, and clearances so the enclosure assembles cleanly and supports internal hardware without extra fasteners or operations.

Mechanism Development & Prototyping

Explored and prototyped multiple actuation layouts to achieve smooth, low-backlash motion suitable for consistent ingredient handling and prep tasks.
Iteratively refined mechanism geometry based on load testing, durability checks, and cycle-to-cycle reliability under real food-prep forces.

Structural & Functional Considerations

Engineered internal structures with predictable force paths and vibration-resistant layouts to maintain accuracy during cutting and handling operations.
Focused on cleanability, accessibility, and food-safe design requirements while keeping components compatible with scalable manufacturing.

🟦 VLA Teleop & Fine-Tuning

Collected high-quality dual-arm teleop demonstrations performing basic food-prep actions to build clean, learnable trajectories.
Used these demonstrations to fine-tune Hugging Face’s SmolVLA model, improving the robot’s ability to interpret visual cues and reproduce manipulation behaviors reliably.

SliceX Prototype built in 1 month

Page updated

Google Sites

Report abuse