Robotics
File size: 1,938 Bytes
b2cfca5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: cc-by-nc-nd-4.0
pipeline_tag: robotics
---

# TCP-Speed

**TCP-Speed** is a baseline model for desired-speed conditioned autonomous driving, introduced as part of the **Bench2Drive-Speed** benchmark.

[**Project Page**](https://thinklab-sjtu.github.io/Bench2Drive-Speed/) | [**Paper**](https://huggingface.co/papers/2603.25672) | [**Github**](https://github.com/Thinklab-SJTU/Bench2Drive-Speed)

## Overview

Bench2Drive-Speed is a closed-loop benchmark for desired-speed conditioned autonomous driving. It enables explicit control over vehicle behavior through:
* 🎯 **Target-speed conditioning**: Users can specify the desired driving speed.
* 🚘 **Overtake / Follow commands**: Users can specify whether the vehicle should overtake or follow others.

The TCP-Speed model serves as a baseline that processes these user preferences to generate driving actions within the CARLA simulator.

## Usage

To use or build upon this agent, refer to the official [Bench2Drive-Speed repository](https://github.com/Thinklab-SJTU/Bench2Drive-Speed). 

The agent inherits from `AutonomousAgent`. In addition to standard CARLA inputs, you can access the user-specified conditions within the agent logic as follows:

```python
# Extract target speed and overtake commands within the agent class
target_speed = self.get_planned_speed() # float (km/h)
do_overtake = self.do_overtake          # boolean
```

For full setup, training, and evaluation instructions, please follow the documentation in the [Github README](https://github.com/Thinklab-SJTU/Bench2Drive-Speed).

## Citation

```bibtex
@article{Bench2DriveSpeed,
    title={Can Users Specify Driving Speed? Bench2Drive-Speed: Benchmark and Baselines for Desired-Speed Conditioned Autonomous Driving}, 
    author={Yuqian Shao and Xiaosong Jia and Langechuan Liu and Junchi Yan},
    year={2026},
    eprint={2603.25672},
    archivePrefix={arXiv},
    primaryClass={cs.RO},
}
```