Perception Infrastructure SDK

Ship the robot,
not the perception stack.

PathVynt is a drop-in sensor-fusion and path-planning stack for robotics teams. LiDAR-camera alignment, dynamic obstacle prediction, and lane-level localization — production-ready from day one.

Request SDK Access See the Architecture
3D LiDAR point cloud scan of an industrial corridor rendered in teal-blue gradient, showing sparse depth data typical of autonomous navigation environments
<20ms Fusion Latency
3 Sensor Modalities
4+ Robot Platforms
2021 Founded, Mountain View CA

Core Capabilities

Everything your robot needs to perceive and move

Four modules covering the full perception-to-motion pipeline. Drop PathVynt in and focus on your robot's mission.

LiDAR-Camera Sensor Fusion

Kalman-based fusion with runtime extrinsic calibration correction. Handles temporal misalignment between LiDAR and camera streams without manual re-calibration.

Dynamic Obstacle Prediction at 20Hz

Velocity estimation and trajectory forecasting using CVTR and social force models. Risk-scored outputs ready for your planner's cost function at 20Hz update rate.

Lane-Level Localization, GPS-Denied

Map-relative localization using LiDAR keyframe matching and particle filter update. Maintains lane-level accuracy in basements, tunnels, and RF-shielded environments.

SDK Architecture

Four modules, one coherent pipeline

PathVynt's modules share a common data model — no serialization cost between stages, no glue code.

LiDAR + Camera Sensor Input 3 modalities Fusion Engine Kalman Alignment <20ms latency Occupancy Grid Dynamic Update 2cm–25cm res Path Planning Motion Output ROS 2 / C++ / Python
Explore the Platform Read the Docs

Robot Platforms

One SDK, multiple deployment contexts

PathVynt adapts to your hardware constraints — not the other way around.

Autonomous mobile robot navigating a modern warehouse aisle with perception sensor array visible

Warehouse AMRs

Cold-storage thermal drift, narrow-aisle constraints, forklift prediction — PathVynt handles them by default.

Learn more
Outdoor delivery robot on a sidewalk with perception sensors active

Delivery Robots

GPS-denied navigation on Jetson Orin NX. Pedestrian trajectory prediction with social force model approximation.

Learn more
Autonomous truck on a highway at dusk with LiDAR sensor cluster mounted on cab roof

Autonomous Trucks

Long-range LiDAR management, highway lane localization with worn markings, adverse weather filtering.

Learn more

From the Field

What robotics engineers say

We spent six months building our own calibration pipeline before finding PathVynt. Dropped it in over a weekend and it handled our thermal LiDAR drift in the freezer environment out of the box. That's six months we didn't lose.

AK
A. Kowalczyk Perception Lead at an autonomous cold-chain logistics company

The obstacle prediction module's 20Hz output was exactly what our planner expected. No adapter layer, no buffering tuning. The API surface is small and the documentation is honest about limitations — rare in this space.

MR
M. Reyes Senior Motion Planning Engineer at a last-mile delivery robotics startup

Running on Jetson AGX Orin at under 18ms end-to-end. We tried two other fusion libraries first and both required us to write wrapper layers to handle the covariance output format. PathVynt's output matched our Eigen-based pipeline directly.

SP
S. Prasad Robotics Software Architect at an autonomous forklift company

Get Started

Ready to drop in PathVynt?

90-day evaluation license available. We'll help you run the first fusion pass on your hardware within a day.

Request SDK Access