All systems shown below were trained and deployed within 2 hours. Connect your robots and sensors to our Skill Processing Unit (SPU) and start automating your most complex tasks today.
Straps are deformable: their state after every motion is uncertain, so open-loop fold sequences drift and fail within a few cycles.
EBMs can learn to handle variations in deformables and exhibit precision when threading the strap through the opening. This is done in a closed-loop using continuous visual feedback to correct any misalignments.
Adhesive labels tear or stay stuck when peel angle and speed are fixed. Adhesion strength varies with batch, substrate, and temperature.
EBMs can learn to modulate peel angle, re-try and grip in real time from visual feedback, lifting labels cleanly across adhesion variations.
Fruit and food can arrive in highly unstructured pick bins where they can constantly move and reorient as the robot interacts with them. The conveyor is also unstructured, and parts can only be placed between the white markers.
All end-to-end vision decision-making, from picking the bottle to placing it on the conveyor. The conveyor placement task was trained with just 2 minutes of data and the pick task was trained with 1 hour of data. The Sydekick Platform makes it easy to orchestrate various skills into one autonomous workflow.
Small, near-identical washers arrive jumbled and overlapping. Sub-millimetre size differences and reflective surfaces make it hard to easily see the parts.
With just 1 hour of data, our EBMs can learn to separate, pick and sort the differnt washer sizes into their respective compartments. No seperate classifier or vision system needed, fully end-to-end.
Long-horizon, dexterous manipluation task to assemble a generic storage crate from Kmart. Requires dual robot arm coordination and memory.
EBMs can orchestrate multiple robot arms with multiple cameras to take on the task with human-level dexterity and adaptability as the crate moves across the table. Attention to detail when clipping the corners and retries if it fails.
Kits combine parts of different shape, weight, and finish into a fixed tray layout. Every SKU change traditionally means new fixtures and re-taught poses.
One EBM skill generalises across the part mix, selecting grasp and placement per item so new kit configurations are a data change rather than a re-programming project.
Fabric wrinkles, slips, and never lands the same way twice. Corner detection and grasping have to work from whatever crumpled state the towel is in.
A learned folding skill finds corners visually, smooths and repositions as needed, and executes the fold sequence without fixtures or fabric-specific tuning.
Bottles arrive in arbitrary orientations with shifting centres of mass. Re-orienting them reliably requires dynamic, non-prehensile motion rather than a static regrasp.
The EBM skill plans and executes the flip in one fluid motion, using visual feedback to correct mid-manoeuvre and land each bottle upright.
Flush, near-zero-thickness cards give a gripper nothing to grab. Too much force damages the card; too little and it slips or lifts two at a time.
A fine-force EBM skill finds the edge, applies just enough pressure to separate a single card, and confirms the pick before moving on.
Long-horizon stacking compounds small errors: one misaligned cup topples the column and stops the cycle.
Sydekick runs the task continuously, re-aligning each cup from live feedback and recovering from disturbances so the stack sustains indefinitely.
