Vision-guided handling
Feeding, stacking, bundling and palletizing across a changing product mix.
Products / End-of-line
Palvex
Physical-AI robotic feeding, stacking and palletizing. Drives robotic feeding, stacking, bundling and palletizing at line speed.
- Palletize rate
- 1.4kper hour
- Cube utilization
- +11%
- Stack topple
- −80%
Why this stage is hard
Finished boxes and cartons must be fed, stacked, bundled and palletized: high-mix, high-speed, physically demanding work that is hard to automate with fixed robotics. Blanks and bundles vary by job, board and count; stacks lean, topple or crush; pallet patterns change constantly.
The physical-AI robotic handling agent
Palvex drives vision-guided feeding, stacking, bundling and palletizing with learned grasp and place policies robust to blank and box variation and stack lean, generating GPU-optimized pallet patterns for stability, cube and due-date, validated sim-first with safety-guarded motion.
- Vision-guided feeding, stacking, bundling and palletizing across product mix
- Learned grasp and place policies robust to blank and box variation and stack lean
- GPU-optimized per-job pallet patterns for stability, cube and due-date
- Sim-first cell validation and changeover with safety-guarded motion
Palvex readouton target
Palletize rate1.4k per hour
Cube utilization+11 %
Stack topple−80 %
Changeover−50 %
What Palvex does
Learned policies
Grasp and place robust to blank and box variation and stack lean.
Optimized patterns
GPU-optimized pallet patterns balance stability, cube and due-date.
Sim-first changeover
Cells validated in simulation before they run, with safety-guarded motion.
Palletize rate
1.4kper hour
Cube utilization
+11%
Stack topple
−80%
Changeover
−50%
How it composes into the plant
Palvex is the end-of-line robotics pillar. It closes the perceive-decide-act loop, taking sound boxes from Gluvex and shipping them stacked, bundled and palletized.
- Shares the plant-edge data flywheel
- Uses cross-vendor machine and MES connectors
- Runs on the common Jetson, Triton and NIM fabric
- Validated by the Twinflo digital twin before every run
Deployment and hardware
| Aspect | Detail |
|---|---|
| Stage | End-of-line |
| Autonomy modes | Shadow, assist, autonomous |
| Perception | Edge vision and sensor fusion at line speed |
| Control | Bounded closed-loop setpoint moves within approved limits |
| Primary NVIDIA hardware | Jetson, Isaac Sim and Lab, DGX |
| Digital twin | Twinflo (Omniverse) pre-run simulation |
−55%Manual handling labor
−80%Stack topple and crush
+11%Pallet cube utilization
−50%Changeover time
High-mix palletizing was our hardest cell to automate. Sim-first changeover made it practical.
The other agents
Board making
Corvex
Autonomous corrugator and board-bond control
Runs the corrugator to hold warp-free, on-strength board across grades and humidity.
Explore Corvex PrintingChromata
Autonomous registration, brand-color and print-defect control
Keeps every impression on-register and on-brand across flexo, litho and digital.
Explore Chromata Die-cut and creaseKreasa
Autonomous die-cut, crease and nesting yield
Maximises board yield and holds clean cuts and sound creases.
Explore KreasaQuestions converters ask
Can we run Palvex on a single line first?
Yes. Palvex deploys on one machine under the Line tier so you can measure a baseline, run in shadow mode, and switch on control as ROI proves out.
Does it work with our existing machine and vendor?
Foldeon is built for mixed fleets. Cross-vendor connectors integrate the machine, its PLC and your MES and OEE systems, with no rip-and-replace.
How do operators stay in control?
You choose the autonomy level per workflow: shadow (recommend), assist (approve each move) or autonomous, and every decision is logged and auditable.
See Palvex run your line
We’ll run Palvex in shadow mode against a real job and show the quality, waste and makeready curves.