Scenarios & Maps¶

RealSim-CP covers three Tokyo environments, each captured under three weather/time conditions.

Maps¶

Map	Environment	RSUs	Sensor-equipped vehicles
Aomi	Complex urban intersections	2–4	1–23
Odaiba / Daiba	Wide roads, multiple intersections	4–8	1–11
Shutoko Expressway	Expressway	2–6	1–47

Each map is captured under clear daytime, rainy daytime, and clear nighttime conditions. The physics-based renderer reproduces the visual character of each condition — note the wet road reflections and reduced visibility at night below.

/// caption Clear daytime (RSU camera) ///

/// caption Clear nighttime (RSU camera) ///

/// caption Rainy daytime (RSU camera) ///

Scenario catalogue¶

Scenario	Status	Variants
Daiba Station	Available (documented below)	sunny ×2, night ×2, rainy ×2
Shutoko	Available	weather/time variants
Aomi Crossing	Planned	TBD

The Daiba Station scenario is the worked example used throughout this documentation.

Daiba Station scenario¶

Modeled on the area around Daiba Station, Tokyo. Six variants across three weather/time conditions, each ~10 s at 10 Hz.

Variant summary¶

Variant	Frames	Image agents	LiDAR agents	RSUs	Vehicles	Label size
`sunny_1`	101	12	8 ⚠	3	9	2.71 MB
`sunny_2`	100	12	0 ⚠	3	9	2.69 MB
`night_1`	101	15	14	4	11	2.71 MB
`night_2`	100	15	14	4	11	2.69 MB
`rainy_1`	101	5	5	4	1	1.31 MB
`rainy_2`	100	5	5	4	1	1.30 MB

⚠ See Known issues for the sunny_1 / sunny_2 point-cloud gaps.

Agent roster — `night_1` (most complete variant)¶

Agent	Role	Cameras	LiDAR
`rsu_1` … `rsu_4`	Road-side units (static)	one each (`camera_1..4`)	one each (`lidar_1..4`)
`vehicle_10000`	Connected vehicle	`camera_9..12`	`lidar_5`
`vehicle_10010`	Connected vehicle	`camera_13..16`	`lidar_6`
`vehicle_10020`	Connected vehicle	`camera_17..20`	`lidar_7`
`vehicle_10030`	Connected vehicle	`camera_21..24`	`lidar_8`
`vehicle_10060`	Connected vehicle	`camera_25..28`	`lidar_9`
`vehicle_10070`	Connected vehicle	`camera_29..32`	`lidar_10`
`vehicle_10080`	Connected vehicle	`camera_33..36`	`lidar_11`
`vehicle_10090`	Connected vehicle	`camera_37..40`	`lidar_12`
`vehicle_10100`	Connected vehicle	`camera_41..44`	`lidar_13`
`vehicle_10190`	Connected vehicle	`camera_45..48`	`lidar_14`
`vehicle_10200`	Connected vehicle	`camera_49..52`	`lidar_15`

Total for night_1: 52 camera streams + 14 LiDAR streams = 66 sensor streams, organized across 86 coordinate systems (1 scene + 15 agent-local + ~70 sensor frames).

Annotation statistics — `night_1`¶

101 frames × 13 tracked objects per frame = 1,313 object instances.
13 unique object IDs persist for the whole clip.

Class	Instances
`TYPE_MEDIUM_CAR`	505
`TYPE_COMPACT_CAR`	303
`TYPE_PEDESTRIAN`	202
`TYPE_SMALL_CAR`	101
`TYPE_BUS`	101
`TYPE_LUXURY_CAR`	101
Total	1,313

Class imbalance¶

Across the full dataset, vehicle classes (especially cars) dominate — the Shutoko Expressway in particular is car-heavy. This imbalance is a known characteristic that affects rare-class detection; see Benchmark.