Demonstrated
Spine surgery
Drilling and decompression next to the nerve root and dura, where a single newton too many causes damage. SmartStencil’s spine demo stops the instrument on nerve root contact and retracts it to safety.
Products
SmartStencil™
- Lane assist for surgical robotics
- The vision-force model for surgical robotics
Surgical robots can see, but they can’t feel. SmartStencil gives any surgical robot a sense of touch: it learns the safe force for every tissue, then keeps the robot in its lane, slowing down, pushing back and retracting before harm is done.
US patent11,631,341 granted, engine PCT filed
1,000sim-to-real trials, hardware agnostic
350Kforce-labelled surgical frames
0new hardware required
Lane assist, liveInstrument approaching a nerve root in spine surgery
Zone
Safe zone
Tip force estimate
0.2N
Normal operation. Force is within the tissue boundary.
- SafeWithin limits
- WarningSlow down
- DeceleratePush back
- Hard stopHalt
- ReturnRetract
Surgical robots can see. They can’t feel.
Force sensing today means sensorised instruments: new hardware, built robot by robot. Most surgical robots don’t have it, so every one of them still relies on a human to prevent harm.
4,500+
adverse events from surgical robots reported to the FDA, 2000–2024
FDA MAUDE database$2.8M
average cost per patient of a surgical complication
JAMA Surgery, 202367%
of surgical errors happen during execution, exactly where robots act
WHO Surgical Safety, 2023Force feedback is locked to hardware
Where force feedback exists, it depends on sensor-equipped instruments. Every other robot is force-blind.
AI models see and act, but don’t feel
Vision-language-action models translate commands into motion. None of them encode how much force is safe for each tissue.
Autonomy can’t be validated without it
Without force awareness, autonomous surgical actions have no safety boundary to prove to regulators.
The missing layer in surgical physical AI
Simulation, synthetic data and robot-motion models are already live. What sits between deciding and doing, knowing what is safe, is not. That is where SmartStencil fits.
Whoever builds the perception and planning, the robot still needs to know what is safe to touch.
Intelligence
Foundation models, reasoning and planning
Mobile assist
Perception, motion and decision-making
Lane assist
SmartStencilConstraint intelligence and force safety: what is safe, for this patient, on this robot
Execution
Robotic hardware, actuation and sensor fusion
Real world
Physical interaction with patient tissue
One force brain for every surgical robot
SmartStencil is a vision-force model. Before the first incision, it builds a patient-specific map of safe force, then feeds it into the robot’s own control and learning stack. It plugs into any VLA or reinforcement-learning pipeline.
What goes in
- Patient anatomy
- CT, MRI and wearable data
- Procedure
- Drilling, suturing, dissection and more
- Robot embodiment
- Any surgical robot, any arm
- Surgeon skill tier
- Expert and trainee, stratified
SmartStencil
Pre-procedure constraint engine
- Patient constraint graph
- Procedure failure map
- Boundary violation model
- Temporal risk sequencing
US Patent 11,631,341 granted. Engine PCT filed.
What comes out
- Patient-specific force envelopes
- Anatomy-specific tissue limits for this patient
- Ranked failure states
- Excess force and boundary violation, ranked by risk
- Constraint-aware policy signal
- Feeds the robot’s VLA and RL reward, so it is constrained before the first incision
Force sensing is optional. Force intelligence isn’t.
SmartStencil works with no new sensors. Where a deployment calls for it, it fuses every signal the operating room offers.
VisionStereo and endoscopic cameras, 1080p–4K
Force and torque6-axis at the instrument tip, 1–4 kHz
AcousticCautery and cutting signatures
PressureTactile contact mapping on graspers
17 ms glass-to-glass latencySub-microsecond sensor syncEdge inference inside the control loop
Lane assist for surgical robotics
Like lane assist in a car, SmartStencil doesn’t take the wheel. It keeps the robot inside safe limits in real time, and it responds the way an expert surgeon would: gradually, not with an emergency brake.
Not a kill switch
The robot self-corrects and retracts back to safety, so the procedure can continue.
Tissue-specific thresholds
Different limits for hard bone, dura and soft tissue, set for each patient.
Supports the surgeon, never replaces them
Every intervention is rendered to the surgeon or robot controller inside the control loop.
Learns from every boundary
Each event becomes a reinforcement-learning signal that improves the robot’s policy.
Safe zone
Normal operation. Force stays within the tissue boundary.
Warning
Approaching the boundary. A slow-down signal is sent to the controller.
Decelerate
Boundary reached. The robot is pushed back with graduated resistance.
Hard stop
Severe breach, such as nerve root contact. Motion halts immediately.
Return to safe zone
The robot retracts on its own. Not just stopped: back on track.
Practise first. Then operate.
The same force limits travel from simulation to the operating room, on the same robot.
1
Train the robot
Reinforcement learning against patient-specific force envelopes, validated across 1,000 sim-to-real trials.
2
Rehearse against the limits
Run the procedure before surgery and see exactly where the boundaries are.
3
Operate on the same robot
The rehearsed constraints go live in the operating room, enforced in real time.
Robots that feel the margin
Robots already have dexterity. SmartStencil gives them geometric intelligence: knowing where the safe edge is, even when the eye can’t see it.
Now exploring
Endometriosis surgery
One in ten women live with endometriosis, and its lesions have no clear margin. Cut too little and it returns; cut too much and organs are injured. Here the no-go zones would be the ureter and bowel.
Built from surgeons’ hands, not scraped from video
SmartStencil’s engine is trained on force, kinematic and contact data captured from expert surgeons over three years on Simulatory’s own biomimetic platforms. The taxonomy of surgical failure behind it can’t be crowd-sourced.
350K
Interaction frames
Force, kinematics and contact streams for each surgical manoeuvre
45+
Expert surgeons
Skill-stratified, multi-institution, multi-procedure
250+
Training sessions
On VRspine Pro and SYNspine biomimetic models
40
Failure-mode events
Annotated with how surgeons recovered from them
3 years
To build the data and the failure taxonomy
A surgeon network and dataset that took years of clinical access to assemble.
Designed to enter the clinic inside your robot
SmartStencil licenses as an embedded layer, so it doesn’t need a regulatory submission of its own. It travels inside the robot maker’s FDA filing and supplies the force-safety validation data that filing needs.
Submissions that integrate SmartStencil are designed to meet the criteria of the FDA’s Safer Technologies Program (STeP), which offers expedited development and review for devices that significantly improve safety.
2a
Reduces known serious adverse events: force boundary violations that cause tissue damage.
2b
Reduces known device failure modes: constraint encoding prevents unsafe robot actions.
Why SmartStencil
No other approach combines surgical interaction data, a protected constraint engine and a layer built to plug into any robot.
| Approach | Surgical interaction data | Works across robots | Regulatory pathway | Physical AI integration |
|---|---|---|---|---|
| SmartStencil | Proprietary biomimetic dataset | Any robot, no new hardware | Embedded in OEM submissions | Vision-force-action layer |
| In-house robot development | Limited external data | Single platform | Long internal pathway | Closed stack |
| Generic robotics safety layers | No surgical data | Partial | No regulatory roadmap | Not designed for surgery |
| Raw reinforcement learning | No proprietary data | Not transferable | No path to FDA | No force constraints |
SmartStencil™
Give your surgical robot a sense of touch
For surgical robot makers, research hospitals and clinical partners. See SmartStencil running on a robot and talk through integration with our team.