Cooperative Perception
Sharing Sensed Objects Over the V2X Link
The core idea behind cooperative perception is that a vehicle's perception range should not be limited by its own line of sight. Each connected vehicle runs its native radar, lidar, and camera pipeline, produces a fused object list, and then transmits a subset of those detections over the 5.9 GHz sidelink. A second vehicle receiving that broadcast can place the remote objects into its own coordinate frame and react to a pedestrian stepping off a curb that its own sensors cannot yet see. This is the decisive difference from cooperative awareness, where each station shares only its own kinematic state. The ETSI Collective Perception Message (CPM) was standardized in TR 103 562 and TS 103 324 precisely to carry these perceived-object containers in a compact ASN.1 UPER encoding.
A single CPM bundles a station data container, a sensor-information container describing the field of view of each onboard sensor, and a set of perceived-object containers. Each perceived object carries a relative position, velocity, acceleration, dimensions, a classification, and a confidence value, typically consuming 35 to 60 bytes per object. Because a busy intersection can produce dozens of objects per vehicle, message inclusion rules and Decentralized Congestion Control (DCC) decide which objects are broadcast and how often, keeping the shared channel below saturation. Receivers must motion-compensate every remote object using its capture timestamp before fusing it, since even 50 ms of latency translates to nearly 2 m of position error at highway speed.
Latency and Time-Synchronization Budget
End-to-end latency dominates whether cooperative perception is safe to act on. The path from remote sensor exposure to local fusion must hold under roughly 100 ms for collision-avoidance use cases, and closer to 50 ms on motorways. Because remote object positions are time-stamped against GNSS-disciplined clocks, all stations need synchronization on the order of 1 ms or better, otherwise the geometric registration of remote objects degrades. C-V2X PC5 Mode 4 and 5G NR sidelink both target sub-20 ms one-way radio latency to fit inside the budget.
Channel Load and Object Inclusion
Collective perception messages are an order of magnitude larger than basic safety messages, so channel congestion, not raw radio range, is the binding constraint. ETSI CPM generation rules suppress an object unless it has moved more than 4 m, changed speed by more than 0.5 m/s, or has gone unreported for over one second, and redundancy mitigation avoids re-broadcasting objects a neighbor already announced. These filters keep the channel busy ratio under the DCC ceiling near 62 percent.
Governing Relationships
Te2e = Tsense + Tencode + Tair + Tdecode + Tfusion ≤ 100 ms
Position staleness error (motion compensation):
Δx ≈ vrel × Te2e (e.g. 36 m/s × 50 ms ≈ 1.8 m)
Offered channel load:
L ≈ Nveh × Rcpm × (Shdr + Nobj × Sobj) × 8 / Cch
Where vrel = relative closing speed, Rcpm = CPM rate (1 to 10 Hz), Nobj = objects per message, Sobj ≈ 35 to 60 bytes, Shdr ≈ 120 bytes, Cch = channel capacity (≈ 6 to 27 Mbit/s on a 10 MHz channel), and L is the resulting channel busy ratio held below ≈ 0.62 by DCC.
Message Type Comparison
| Message | What it shares | Typical size | Rate | Standard | Primary benefit |
|---|---|---|---|---|---|
| CPM (cooperative perception) | List of perceived objects | 500 to 1,200+ bytes | 1 to 10 Hz | ETSI TS 103 324 | See unconnected road users |
| CAM (awareness, EU) | Own state only | ~100 to 300 bytes | 1 to 10 Hz | ETSI EN 302 637-2 | Know connected peers |
| BSM (awareness, US) | Own state only | ~200 to 400 bytes | 10 Hz | SAE J2735 | Know connected peers |
| DENM / EVA | Event / hazard notice | ~100 to 800 bytes | Event-driven | ETSI EN 302 637-3 | Warn of incidents |
| MCM (maneuver) | Planned trajectory | ~300 to 800 bytes | 1 to 10 Hz | ETSI TS 103 561 | Coordinate motion |
Frequently Asked Questions
How does cooperative perception differ from cooperative awareness (CAM/BSM)?
Awareness messages (CAM in Europe, BSM in the US) broadcast only the sender's own position, heading, and speed at 1 to 10 Hz. Cooperative perception instead shares the objects a vehicle has detected with its own radar, lidar, and cameras, carried as perceived-object containers in the ETSI Collective Perception Message. That lets a receiver sense non-connected pedestrians and cars it cannot detect directly. The cost is data volume: a CPM can reach 1,200 bytes or more versus roughly 100 to 300 bytes for a CAM.
What end-to-end latency budget does cooperative perception require?
Total glass-to-glass latency from remote sensor capture to local fusion should stay under about 100 ms, and near 50 ms at highway speed, where a vehicle covers 1.8 m in 50 ms. The budget splits across remote sensing (10 to 30 ms), encoding (1 to 5 ms), the air interface (PC5 sidelink ~1 to 20 ms), and decode plus fusion (5 to 20 ms). C-V2X Mode 4 and 5G NR sidelink target sub-20 ms one-way radio latency, which is why GNSS time sync better than 1 ms is essential.
What channel capacity and congestion control does collective perception need?
CPMs are much larger than safety messages, so channel load dominates scaling. A 10 MHz channel at 5.9 GHz carries roughly 6 to 27 Mbit/s. ETSI generation rules only include an object if it has moved more than 4 m, changed speed by more than 0.5 m/s, or gone unsent for over one second, while Decentralized Congestion Control caps the channel busy ratio near 62 percent and redundancy mitigation avoids re-reporting an object a neighbor already sent.