Occupancy Gridmap
Before a planner can choose a path, it needs to know where the world is solid and where it’s clear. Occupancy Gridmap turns the radar point cloud into a probabilistic map of the static scene: every cell scored as occupied, free, or not yet observed, and sharpened over time using radar odometry. It’s the structured spatial layer that Radar Freespace and path planning are built on.
Probabilistic
occupancy per cell
Planner-ready
spatial foundation
Cell-based
occupied / free
On-radar
built on the radar SoC
Applications
From radar returns to a map of free and occupied space.
Automotive
ADAS and autonomous driving need a dense, reliable picture of the static scene — kerbs, barriers, parked vehicles, road edges. Occupancy Gridmap provides it, and it’s what Radar Freespace clusters into the drivable boundary the planner steers to.
Industrial
Mining, agriculture, construction — unstructured sites with no lane to follow. A probabilistic map of fixed structures and obstacles gives autonomous machinery a reliable sense of what is solid and what is traversable.
Indoor / GPS-denied
Multi-storey car parks, warehouses, tunnels. Built from radar and accumulated with Radar Odometry, the occupancy map works with no satellite signal — mapping occupied and free space where GNSS is unavailable.
How it works
From radar point cloud to a probabilistic occupancy map.
STEP 01
Start from the radar point cloud
Each frame, the radar returns a sparse cloud of reflections from the world around it. Returns from moving objects are removed using the radar’s Doppler velocity, so what remains is the static scene — kerbs, walls, barriers, parked vehicles, road edges.
STEP 02
Accumulate it over time
Radar Odometry measures exactly how the radar moved between frames, so successive point clouds can be aligned and overlaid. Sparse single-frame data builds into a dense, stable picture of the surroundings.
STEP 03
Score every cell for occupancy
The space around the vehicle is divided into a grid of cells. Each cell carries a probability — occupied, free, or not yet observed — updated as evidence accumulates, so confidence grows the longer something is seen. The result is a heatmap of where the static world is and where the clear space is.
Integration
Microservice, edge-resident
Occupancy Gridmap ships inside Provizio’s production radar software, as a microservice in the on-radar perception stack. It takes in the radar point cloud and the ego-motion estimate from Radar Odometry, builds the occupancy grid on-device, and publishes it over DDS — where other perception modules like Radar Freespace pick it up. No external compute box required; each microservice can be updated independently, over the air.
- Runs on the radar’s automotive SoC
- Built from the radar point cloud + Radar Odometry
- Occupied / free / unknown probability per cell
- DDS occupancy-grid output
- Independent over-the-air updates per microservice
Questions, answered
Occupancy Gridmap FAQ
They’re two stages of one pipeline, not rivals. Occupancy Gridmap is the underlying probabilistic map — every cell scored occupied, free, or unknown. Radar Freespace is derived from it: the occupied cells are clustered into a single drivable polygon. Take the full occupancy grid when you want the rich probabilistic picture for planning or your own logic; take Radar Freespace when you just need the lightweight drivable boundary. Many platforms ship both.
No — by design it maps the static world. Returns from moving objects are removed using the radar’s Doppler velocity and the ego-motion estimate, so the accumulated map is a stable picture of fixed structure: kerbs, walls, barriers, parked vehicles. That is exactly what freespace and planning need. Moving objects are handled separately by Tracking.
On the radar’s own automotive SoC, as a microservice in the on-radar perception stack. It takes in the radar point cloud and Radar Odometry, builds the grid on-device, and publishes it over DDS at the sensor’s frame rate. No external compute box required.
The map your planner builds on.
If your stack needs a dense, probabilistic picture of the static world — the foundation Radar Freespace and planning build on — Occupancy Gridmap delivers it, on the radar.