# opendrivelab/uniad

**Attribution required: if you use, quote, or summarise this content, you must credit and link back to [awesome-repositories.com](https://awesome-repositories.com/repository/opendrivelab-uniad).**

4,645 stars · 548 forks · Python · Apache-2.0

## Links

- GitHub: https://github.com/OpenDriveLab/UniAD
- awesome-repositories: https://awesome-repositories.com/repository/opendrivelab-uniad.md

## Topics

`autonomous-driving` `autonomous-driving-framework` `bev-segmentation` `end-to-end-autonomous-driving` `motion-planning` `motion-prediction` `multi-object-tracking` `occupancy-prediction` `perception-prediction-planning`

## Description

UniAD is a unified deep learning framework for autonomous driving that integrates perception, prediction, and planning into a single end-to-end model. It functions as a neural network architecture that maps raw sensor data directly to driving trajectories and motion plans.

This project serves as a research implementation of a planning-oriented approach that jointly trains occupancy, mapping, and object tracking modules. It employs a multi-task perception framework to optimize overall driving performance.

The system covers a broad capability surface including end-to-end driving pipelines, vehicle motion optimization, and visual feature aggregation. It coordinates various autonomous driving tasks to refine the entire driving process in a single training cycle.

## Tags

### Artificial Intelligence & ML

- [End-to-End Driving Pipelines](https://awesome-repositories.com/f/artificial-intelligence-ml/end-to-end-architectures/end-to-end-driving-pipelines.md) — Tune track, map, motion, and occupancy modules together to refine the entire driving process in one training cycle. ([source](https://cdn.jsdelivr.net/gh/opendrivelab/uniad@v2.0/README.md))
- [End-to-End Architectures](https://awesome-repositories.com/f/artificial-intelligence-ml/end-to-end-architectures.md) — Implements a neural network that maps raw sensor data directly to driving actions without intermediate pipelines.
- [Multi-Task Vision Training](https://awesome-repositories.com/f/artificial-intelligence-ml/model-training-frameworks/vision-model-training/multi-task-vision-training.md) — Jointly trains occupancy, mapping, and object tracking modules within a single unified vision model.
- [Perception Module Training](https://awesome-repositories.com/f/artificial-intelligence-ml/large-scale-training-frameworks/perception-dataset-processors/perception-module-training.md) — Creates stable weights for tracking and mapping by aggregating visual features across multiple video frames. ([source](https://cdn.jsdelivr.net/gh/opendrivelab/uniad@v2.0/README.md))
- [Temporal Feature Aggregation](https://awesome-repositories.com/f/artificial-intelligence-ml/model-weight-management/weight-distribution/weight-aggregation-configurators/temporal-feature-aggregation.md) — Combines visual information across multiple video frames to create stable representations for tracking and mapping.
- [Planning-Oriented Learning](https://awesome-repositories.com/f/artificial-intelligence-ml/planning-oriented-learning.md) — Directs training of perception modules using the final driving trajectory as the primary signal for gradient updates.

### Hardware & IoT

- [Autonomous Driving Stacks](https://awesome-repositories.com/f/hardware-iot/integration-performance/automotive-software-systems/autonomous-driving-stacks.md) — Provides a comprehensive software stack integrating perception, prediction, and planning for autonomous vehicle operation. ([source](https://cdn.jsdelivr.net/gh/opendrivelab/uniad@v2.0/README.md))

### Part of an Awesome List

- [Motion Planning](https://awesome-repositories.com/f/awesome-lists/devtools/motion-planning.md) — Implements trajectory optimization and path planning by refining track, map, and occupancy modules.

### Data & Databases

- [Occupancy-Based Motion Prediction](https://awesome-repositories.com/f/data-databases/data-processing-pipelines/data-transformation/array-tensor-manipulation/array-filtering/grid-generation/occupancy-grid-generators/probabilistic-occupancy-grids/occupancy-based-motion-prediction.md) — Uses volumetric grid representations to predict the movement of surrounding agents and identify navigable space.

### Software Engineering & Architecture

- [Multi-Task Joint Training](https://awesome-repositories.com/f/software-engineering-architecture/pipeline-optimization-techniques/multi-objective-optimization/multi-task-joint-training.md) — Trains perception and planning modules simultaneously to optimize visual features for downstream decision making.