Robotic portioning and assembly of deformable food components into meal trays
This task involves picking, scooping, and depositing measured portions of varied food components (rice, noodles, proteins, vegetables, sauces, bread rolls, garnishes) into airline and institutional meal trays/casseroles on a high-mix production line, downstream of large-batch cooking and upstream of sealing/packing. The objects are deformable, fragile, and highly variable in geometry, weight, moisture, and consistency, and portions must hit weight/appearance specs without crushing or smearing. SATS has explicitly designated this previously-manual meal-assembly step for automation via robotic finger grippers and auto-dispensing units at its S$150M Jurong Food Hub, which targets capacity of up to ~200,000 packed meals per day. It is hard for a robot because each food class behaves differently, portions are randomly arranged and visually self-similar, and success depends on modulating force per item rather than following a fixed trajectory. Food safety/hygiene and portion accuracy add constraints that pure vision-and-pick systems struggle to meet. We identified this through our own research; we have not confirmed the specifics with the customer directly. This page is our researched read — a starting point for that conversation.
What the task is
RESEARCHED · our reconstructionThis task involves picking, scooping, and depositing measured portions of varied food components (rice, noodles, proteins, vegetables, sauces, bread rolls, garnishes) into airline and institutional meal trays/casseroles on a high-mix production line, downstream of large-batch cooking and upstream of sealing/packing. The objects are deformable, fragile, and highly variable in geometry, weight, moisture, and consistency, and portions must hit weight/appearance specs without crushing or smearing. SATS has explicitly designated this previously-manual meal-assembly step for automation via robotic finger grippers and auto-dispensing units at its S$150M Jurong Food Hub, which targets capacity of up to ~200,000 packed meals per day. It is hard for a robot because each food class behaves differently, portions are randomly arranged and visually self-similar, and success depends on modulating force per item rather than following a fixed trajectory. Food safety/hygiene and portion accuracy add constraints that pure vision-and-pick systems struggle to meet.
Is this the actual task and sequence? What are the real tolerances, cycle rate, and reject criteria, and which steps are today's manual bottleneck? Answering these is what turns this from a researched signal into a validated use case.