Datasets:

Modalities:
Video
ArXiv:
Libraries:
Datasets
License:
File size: 11,972 Bytes
409f152
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d1f9d48
409f152
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
---
license: apache-2.0
---

# ACT-Bench

`ACT-Bench` is a dedicated framework for quantitatively evaluating the action controllability of world models for autonomous driving.
It focuses on measuring how well a world model can generate driving scenes conditioned on specified trajectories.


<img src="https://github.com/turingmotors/ACT-Bench/blob/main/assets/overview.png" alt="Overview of the Dataset" width="600">

For more details, please refer our [paper]() and [repository](https://github.com/turingmotors/ACT-Bench).


## Data fields

| **Key**                       | **Value**                                                        |
|-------------------------------|------------------------------------------------------------------|
| **sample_id**                 | 0                                                                |
| **label**                     | 'straight_constant_speed/straight_constant_speed_10kmph'         |
| **context_frames**            | ['sweeps/CAM_FRONT/n015-2018-08-02-17-16-37+0800__CAM_FRONT__1533201487512460.jpg','sweeps/CAM_FRONT/n015-2018-08-02-17-16-37+0800__CAM_FRONT__1533201487612460.jpg','samples/CAM_FRONT/n015-2018-08-02-17-16-37+0800__CAM_FRONT__1533201487762460.jpg'] |
| **instruction_trajs**         | [[[0.0, 0.0, 0.0], [0.1545938742, 0.0015977411, -0.0206596931], [0.3063817081, 0.0019410192, -0.0114837886], [0.4497439065, 0.0027015515, -0.027025583], [0.6050902104, 0.002897561, -0.0230843033], [0.7498937661, 0.0028746845, -0.0432883387], [0.8932801666, 0.0027740429, -0.0449931452], [1.0461783792, 0.0027341864, -0.0609023298], [1.1934560378, 0.0026207325, -0.0626793795], [1.3314688069, 0.002331042, -0.083836065], [1.4882952761, 0.0021888225, -0.0833974201], [1.6343445922, 0.0021784357, -0.1030258874], [1.7774288686, 0.0019280693, -0.0995250479], [1.9282369453, 0.0017355272, -0.1257697433], [2.0736730734, 0.0013283469, -0.1290765928], [2.2130084402, 0.0011821295, -0.1515462308], [2.3587170349, 0.0011464657, -0.1489038888], [2.5127366379, 0.0010401979, -0.1685882206], [2.652663411, 0.0008351443, -0.1706014231], [2.8040034852, 0.0005308638, -0.1906429445], [2.9546874643, 0.0003028058, -0.1814105658], [3.098129893, 0.0001099507, -0.1986876182], [3.2477339776, -9.86779e-05, -0.1938415363], [3.3913945285, -0.0004952867, -0.2175208151], [3.5375306412, -0.0010135945, -0.2182340147], [3.6820731288, -0.001606249, -0.2416164848], [3.8279886236, -0.0021923962, -0.2411775227], [3.969924299, -0.0025448799, -0.2629197723], [4.1173996536, -0.0032625234, -0.263342105], [4.2608852146, -0.00372057, -0.2862758575], [4.3976864233, -0.0043610743, -0.2868744325], [4.5461465324, -0.0048756002, -0.3147401786], [4.6937375295, -0.0055456191, -0.3118187509], [4.8355738212, -0.0058713778, -0.3335816396], [4.9815369191, -0.0058726867, -0.3481201454], [5.1292536114, -0.0065586828, -0.343004249], [5.2652689873, -0.0072471006, -0.3474833218], [5.4155127607, -0.0074426697, -0.3684240186], [5.5638769338, -0.0081954509, -0.3638649342], [5.707405646, -0.0085145329, -0.37765957], [5.8565373943, -0.0093398237, -0.3754173488], [5.9987280205, -0.0099226852, -0.4002108294], [6.1446056388, -0.0107350954, -0.4018748844], [6.2867674027, -0.0115938312, -0.4275775659], [6.4344388492, -0.0125163437, -0.4219962191], [6.576710747, -0.0136388196, -0.4450902563], [6.716435109, -0.0145731886, -0.4416513665], [6.868338088, -0.0157876493, -0.4588417966], [7.0145481629, -0.0169398894, -0.4566243329], [7.1504452338, -0.0183231311, -0.4806745948], [7.3029298241, -0.0194984322, -0.4857886661], [7.4431224522, -0.0208558857, -0.5107711508], [7.5846788069, -0.0219164955, -0.5117771397], [7.7352918213, -0.0229614355, -0.5298967143], [7.8822503429, -0.0238655488, -0.5281344161], [8.0203600833, -0.0247095883, -0.5483177376], [8.1746536442, -0.0259923694, -0.5476485202], [8.3163978205, -0.0268716349, -0.5702512244], [8.4553645875, -0.0278297602, -0.5790391197], [8.5969749414, -0.0289897489, -0.6055032887]], ... |
| **reference_traj**            | [[0.0, 0.0, 0.0], [0.3063817081, 0.0019410192, -0.0114837886], [0.6050902104, 0.002897561, -0.0230843033], [0.8932801666, 0.0027740429, -0.0449931452], [1.1934560378, 0.0026207325, -0.0626793795], [1.4882952761, 0.0021888225, -0.0833974201], [1.7774288686, 0.0019280693, -0.0995250479], [2.0736730734, 0.0013283469, -0.1290765928], [2.3587170349, 0.0011464657, -0.1489038888], [2.652663411, 0.0008351443, -0.1706014231], [2.9546874643, 0.0003028058, -0.1814105658], [3.2477339776, -9.86779e-05, -0.1938415363], [3.5375306412, -0.0010135945, -0.2182340147], [3.8279886236, -0.0021923962, -0.2411775227], [4.1173996536, -0.0032625234, -0.263342105], [4.3976864233, -0.0043610743, -0.2868744325], [4.6937375295, -0.0055456191, -0.3118187509], [4.9815369191, -0.0058726867, -0.3481201454], [5.2652689873, -0.0072471006, -0.3474833218], [5.5638769338, -0.0081954509, -0.3638649342], [5.8565373943, -0.0093398237, -0.3754173488], [6.1446056388, -0.0107350954, -0.4018748844], [6.4344388492, -0.0125163437, -0.4219962191], [6.716435109, -0.0145731886, -0.4416513665], [7.0145481629, -0.0169398894, -0.4566243329], [7.3029298241, -0.0194984322, -0.4857886661], [7.5846788069, -0.0219164955, -0.5117771397], [7.8822503429, -0.0238655488, -0.5281344161], [8.1746536442, -0.0259923694, -0.5476485202], [8.4553645875, -0.0278297602, -0.5790391197], [8.7460786149, -0.0302324411, -0.6148562878], [9.040228578, -0.0320762238, -0.6391508753], [9.3238627154, -0.0334427094, -0.6567384988], [9.6242967538, -0.0349175272, -0.675390711], [9.9159747274, -0.0361669985, -0.7020474284], [10.2029848123, -0.0383206259, -0.7409547588], [10.485217797, -0.0402655886, -0.7784671144], [10.7852857398, -0.0415422365, -0.808403356], [11.0714450976, -0.0426406971, -0.8327939143], [11.3716683909, -0.0438444619, -0.8601736098], [11.6663477515, -0.044536854, -0.8800681964], [11.9537060995, -0.0457889104, -0.9064147281], [12.2546047035, -0.046582522, -0.932251343], [12.5430745076, -0.046996187, -0.961586981], [12.8331523584, -0.0482537294, -0.9948334053], [13.1342964502, -0.0489972987, -1.0360002826], [13.4312132278, -0.0493575167, -1.0671797101], [13.7167782768, -0.0493845646, -1.0965326758], [14.0242449794, -0.0484199283, -1.122908192], [14.3217588305, -0.0482550798, -1.1533763216]] |
| **intrinsic**                 | [[1266.4172030466, 0.0, 816.2670197448],[0.0, 1266.4172030466, 491.5070657929],[0.0, 0.0, 1.0]] |


### Labels

The `label` field is a string that represents one of the following high-level driving actions:

```python
[
    'curving_to_left/curving_to_left_moderate',
    'curving_to_left/curving_to_left_sharp',
    'curving_to_left/curving_to_left_wide',
    'curving_to_right/curving_to_right_moderate',
    'curving_to_right/curving_to_right_sharp',
    'curving_to_right/curving_to_right_wide',
    'shifting_towards_left/shifting_towards_left_short',
    'shifting_towards_right/shifting_towards_right_long',
    'shifting_towards_right/shifting_towards_right_short',
    'starting/starting_20kmph',
    'starting/starting_25kmph',
    'starting/starting_30kmph',
    'stopping/stopping_15kmph',
    'stopping/stopping_20kmph',
    'stopping/stopping_25kmph',
    'stopping/stopping_30kmph',
    'stopping/stopping_35kmph',
    'stopping/stopping_40kmph',
    'stopping/stopping_45kmph',
    'straight_accelerating/straight_accelerating_15kmph',
    'straight_accelerating/straight_accelerating_20kmph',
    'straight_accelerating/straight_accelerating_25kmph',
    'straight_constant_speed/straight_constant_speed_10kmph',
    'straight_constant_speed/straight_constant_speed_15kmph',
    'straight_constant_speed/straight_constant_speed_20kmph',
    'straight_constant_speed/straight_constant_speed_25kmph',
    'straight_constant_speed/straight_constant_speed_30kmph',
    'straight_constant_speed/straight_constant_speed_35kmph',
    'straight_constant_speed/straight_constant_speed_40kmph',
    'straight_constant_speed/straight_constant_speed_45kmph',
    'straight_constant_speed/straight_constant_speed_5kmph',
    'straight_decelerating/straight_decelerating_30kmph',
    'straight_decelerating/straight_decelerating_35kmph',
    'straight_decelerating/straight_decelerating_40kmph'
]
```

The [preprocessing function](https://github.com/turingmotors/ACT-Bench/blob/main/act_bench/compute_score.py#L92) in the `ACT-Bench` converts the above labels into the following 9 classes to be compared with the estimated action by the `ACT-Estimator`.


### Context Frames

The `context_frames` field contains the list of image paths that are used to generate the driving scenes conditioned on the `instruction_trajs`. This image path is relative to the `dataroot` directory of the nuScenes dataset. Make sure to download the [nuScenes](https://www.nuscenes.org/nuscenes) dataset before generating the driving scenes.


### Command Classes

The `command` output by the `ACT-Estimator` represents the predicted high-level driving action.
The output is a logits of 9 classes as follows:

```python
LABELS = [
    "curving_to_left",              # command = 0
    "curving_to_right",             # command = 1
    "straight_constant_high_speed", # command = 2
    "straight_constant_low_speed",  # command = 3
    "straight_accelerating",        # command = 4
    "straight_decelerating",        # command = 5
    "starting",                     # command = 6
    "stopping",                     # command = 7
    "stopped",                      # command = 8
]
```

### Instruction Trajectory and Reference Trajectory

The `instruction_trajs` (shaped as `(50, 60, 3)`) is created by dividing the `reference_traj` into segments at each time step, and serves as the input to condition the driving scene generation. The `reference_traj` (shaped as `(50, 3)`) is the ground truth trajectory to evaluate the performance of action controllability of the driving world model. So it is expected that if the generated video accurately follows the `instruction_trajs`, the generated trajectory should be close to the `reference_traj`.
Although both consists of 50 waypoints at each time step, the ACT-Bench evaluation framework only uses the first 44 waypoints to evaluate the performance of action controllability of the driving world model.

Each waypoint is represented as a 2D vector `(x, y)` in a 2D Cartesian coordinate system.

- The origin `(0, 0)` is defined as the initial position of the vehicle at the start of the video.
- The `x`-axis corresponds to the forward direction of the vehicle, with positive values indicating forward movement.
- The `y`-axis corresponds to the lateral direction of the vehicle, with positive values indicating movement to the left.

Note that this coordinate system is different from the one used in the ACT-Estimator's `waypoints` output. The conversion between the two coordinate systems is automatically performed by the `ACT-Bench` evaluation framework.


## Authors

Here are the team members who contributed to the development of `ACT-Bench`:

- Hidehisa Arai
- Keishi Ishihara
- Tsubasa Takahashi
- Yu Yamaguchi


## How to use

The following code snippet demonstrates how to load the `ACT-Bench` dataset.

```python
from datasets import load_dataset

benchmark_dataset = load_dataset("turing-motors/ACT-Bench", data_files="act_bench.jsonl", split="train")
```

See [here](https://github.com/turingmotors/ACT-Bench/) for the instructions on how to evaluate driving world models using `ACT-Bench`.

## License

The ACT-Bench is licensed under the Apache License 2.0.


## Citation

If you find our work helpful, please feel free to cite us.

```bibtex
@misc{arai2024actbenchactioncontrollableworld,
      title={ACT-Bench: Towards Action Controllable World Models for Autonomous Driving},
      author={Hidehisa Arai and Keishi Ishihara and Tsubasa Takahashi and Yu Yamaguchi},
      year={2024},
      eprint={2412.05337},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2412.05337},
}
```