November 14, 2019      

Voyage, which is developing self-driving vehicles, today announced the public release of Voyage Deepdrive, a free and open-source self-driving car simulator. In addition, the company announced hiring Craig Quiter, a long-time open source developer in the industry and creator of, which Voyage’s platform is based.

In a blog post announcing the platform, Voyage CTO Drew Gray said there is an increasing need to build new data-driven algorithms that can handle the complexity of driving, and where heuristic rule-based algorithms fall short. He said there are “countless engineers who could be positioned to help tackle these challenges,” but this often requires having access to the right tools or a vehicle, which is limited to the engineers working at self-driving car companies.

“The story of Deepdrive goes back quite a few years when Craig – inspired by the progress made in AI and the positive impact a self-driving car could have on society – started the project as an open-source platform to enable anyone in the world to work on self-driving technology,” said Gray.

Voyage Deepdrive simulation software

Image: Voyage

Sandbox for exploration

He said the heart of Deepdrive is a focus on end-to-end learning and deep reinforcement learning. The company works with Applied Intuition to drive its core production software forward, but said that Deepdrive will give them a sandbox for research and exploring academic approaches. “Deepdrive’s first class support for reinforcement learning means that researchers can focus solely on the problem of driving rather than worrying about the stack required to run their agents on a physical car,” said Gray.

“By launching Voyage Deepdrive, we want to democratize self-driving car research by decoupling the algorithm development from the hardware,” Gray continued. “Our vision is to provide these tools to anyone who could benefit from them, while ensuring that their work can really benefit and progress the field.”

The company plans to achieve this by having requirements from real self-driving car companies that face these challenges every day, building worlds, scenarios, and events that have been historically hard to solve, Gray said. In addition, the company wants to build an experimental sandbox to explore deep reinforcement learning and other machine learning approaches that can move the technology forward, as well as reduce the barriers to entry so they can attract more engineers to the field.

Features of the platform include:

  • Easy access to sensor data, with interfaces to grab camera, depth and vehicle data to build and train models.
  • Three diverse maps, including a 25-block San Francisco-style map in addition to an open and modifiable five-block and cityscape loop.
  • Advanced in-game AI, with agents that overtake, follow, route, and negotiate intersections intelligently.
  • Deep reinforcement learning PPO2 baseline agent and end-to-end MNET2 imitation learning baseline.
  • Python Unreal scripting, giving access to the entire Unreal API in Python. Using UnrealEnginePython, the system can read any property and call any method on every actor in the scene.
  • Domain randomization, which gives seven view modes to support transfer learnining. These include depth, world normal, ambient occlusion, base color, roughness, and reflectivity.

The company said it has launched an ongoing leaderboard, where teams and individuals can see how their agent stacks up against others. In addition, the simulator features an accurate 3D world built by Parallel Domain, giving developers an environment that allows code testing to be more realistic.

Voyage said it will be hosting a series of competitions in the coming months to encourage independent engineers and researchers to identify AI solutions to specific scenarios and challenges that actual self-driving cars face on the roads. Engineers looking for more information on the platform should visit the Voyage Deepdrive website.