AI robot PoC and field trialsNarrow it to one process and get the answer on a real robot.
Whether a robot can be deployed is not settled by a catalogue. Pick one task, run it on the actual objects, and measure success rate and takt. EmplifAI takes on process selection, arranging the machine, designing the evaluation, and running the trial on a real robot.
Discuss a PoC
Factory/warehouse transportation simulation
Unitree G1
In an Isaac Sim physics simulation, G1 held a 0.25 kg box with both hands, walked approximately 1.2 m, and placed it on another table. This is simulated footage.
RealMan RMC-AIDAL
We grasped the handle of a hypothetical transport case with a body size of 30 x 20 x 20 cm and a total weight of 10 kg, and verified the process of transporting it on wheels for approximately 1.2 m, placing it on a stand, and then releasing it. This is a physical simulation that corrects the center of gravity and contact settings of the public CAD model, and does not guarantee the load capacity of the robot.
RealMan RMC-AIDAL: 10 kg case transport with both arms
We verified the operation of holding a hypothetical transport case with a body size of 30 x 40 x 20 cm and a total weight of 10 kg with both hands using the left and right handles, transporting it approximately 1.2 m, and placing it on a stand. During transportation, contact between the left and right hands and load sharing are recorded. This is a physical simulation that corrects the center of gravity and contact settings of the public CAD model, and does not guarantee the load capacity of the robot.
You can also consult with us about simulations and demonstration experiments tailored to your business, such as replenishing shelves, supplying parts boxes, and sorting.
Environment/box 3D materials: NVIDIA SimReady / codepawl/simready-glb · CC BY 4.0 · Assets were converted, resized, and assigned adjusted surface materials.
A PoC fails when its scope is too wide
When a robot PoC goes wrong, the cause is usually not the technology. It is that the scope is too wide. Start from "we want to verify factory automation" and nothing settles which process to run or what to measure, so it ends without a decision.
A PoC that produces something has its target narrowed to one. This part, from this position, into this jig. What share succeeds, how many seconds each time. Make it that concrete and the real machine gives you an answer.
And the point of a PoC is not to succeed but to decide. Learning that something does not work has the same value. Once you know whether a change of conditions would make it work, or whether it is out of reach for now, you can make the next investment decision.
Hardware Verification: Gemini 4 Argon × Unitree G1 Plastic Bottle Grasping
At our lab, Unitree G1 picked up a plastic bottle using Gemini 4 Argon. No hard-coded motions. Just AI turning camera vision directly into action. The future of robotics is getting real.
Even when the object's position changes, the AI visually perceives the scene and generates the grasping motion. We undertake custom PoC and verification projects that integrate frontier AI models and VLAs onto real hardware to evaluate feasibility on your specific objects and workflows.
How a PoC runs
You can talk to us before the machine is chosen, and before the process is narrowed.
- 01 Process selection
- We pick the first process to verify from the variation in the objects, the slack in the takt, and the burden on people. This matters most.
- 02 Design the evaluation criteria
- We agree before starting on what counts as success — rate, takt, time to recover, and so on.
- 03 Arrange the robot and configuration
- We consider the payload, reach and hand required, then arrange the machine on rental. You can try before buying.
- 04 Verify on the real robot
- We run it on the actual site, or at our premises using your objects. Conditions are varied to identify where it holds.
- 05 Report
- We organise the measurements, the conditions that held and the conditions that did not, and leave them in a form usable for the next decision.
- 06 On to the next scope
- Using the process that worked as the baseline, we consider adjacent processes and adding machines.
Ways to approach the validation
Run it under remote operation first
A person operates it and the site runs before full autonomy, so you get the working conditions and the training data at once.
Narrow it down in simulation
Settle the conditions in NVIDIA Isaac Sim and Isaac Lab before moving to the real robot, so the line stops for less time.
Obtain the motion by learning
For motion that cannot be written out, we consider learning it from human demonstration.
Rent the machine
Try it on the target process before buying. If it does not fit, you can switch to another machine.
Gather the data
We take on everything from designing the conditions for the training data through collection and delivery.
How it runs in a factory
Transport, picking, jig loading and other target tasks on the manufacturing floor are gathered here.
NVIDIA GEAR Sonic × Unitree G1: Real robot demonstration
We ran NVIDIA GEAR Sonic on a Unitree G1 robot and verified its operation. Watch the video to see the demonstration.
Gemini 4 Argon × Unitree G1: Turning Vision into Action
Unitree G1 just picked up a plastic bottle using Gemini 4 Argon. No hard-coded motions. Just AI turning vision into action. The future of robotics is getting real.
Even when the bottle is moved and placed again, the AI visually understands the scene and generates the grasping motion. We provide Physical AI and VLA implementation on the Unitree G1 as well as custom PoC and verification development tailored to your operations.
Frequently asked questions
I am not sure which process to try first.
Process selection can be discussed with us. Processes with little variation in the objects, slack in the takt and a heavy burden on people suit best. We hear the situation on site and narrow the candidates.
How much does a PoC cost and how long does it take?
It changes with the difficulty of the task, whether a machine is already available, and how far the evaluation goes. We first organise the goal and the criteria, then quote the scope, cost and schedule per stage.
Can we validate without stopping our own line?
One option is to start with verification in simulation. The final check happens on a real robot, but the time the line is occupied gets shorter. We can also take custody of the objects and verify at our site.
What happens if it does not work out?
Learning that it does not work is itself the result of a PoC. We put in the report which conditions failed and what could change that. We will not push a deployment that does not fit.
Discussing a PoC or field trial
Tell us the target process, the objects handled and the current work. A stage where the process is not yet narrowed is fine. We will propose it as a trial starting from one process.
Discuss a PoC