Waymo Unveils Waymax, the Next-Gen Lightweight Simulator for Autonomous Driving Research 

Oct. 19, 2023 -- Waymo announced today the launch of Waymax, a groundbreaking lightweight simulator tailor-made for the progression of autonomous driving research.

Image: Waymo

Building on over a decade of spearheading advances in artificial intelligence, machine learning, autonomous driving, robotics, and computer vision, Waymo has consistently shared its knowledge with the research community. This has been evident through their contributions like the Waymo Open Dataset initiative and presentations at prominent academic conferences. With Waymax, Waymo ushers in a fresh chapter in its collaborative endeavors.

Simulation plays an instrumental role in evaluating the performance of autonomous driving systems. Waymo's dedication to this facet is evident in their massive investments in hyper-realistic simulation environments, such as Simulation City, which assess the Waymo Driver's capabilities.

Effective simulators rely on extensive, meticulously organized data sets mirroring real-world scenarios. It's imperative not just for the simulation to visually emulate reality, but also for entities within the simulation—like human drivers, pedestrians, and cyclists—to mimic real-world behavior. Consequently, the interaction of these simulated agents and their link to planning research has grown in significance.

Leveraging data from tens of millions of real-world miles amassed by the Waymo Driver, Waymo's in-house simulators are already a force to be reckoned with. This new release promises to share these unparalleled capabilities with the wider academic community, marking Waymax as a pioneering, lightweight simulator.

Waymax stands apart from its contemporaries, offering unique features and functionalities. Some highlights include:

• Lightweight Design: Ensuring efficient experimentation and swift iterations for researchers.
• Data-Driven: It harnesses over 570 hours of real-world driving data from various cities, available in the Waymo Open Motion Dataset, to emulate realistic driving patterns.
• Hardware Optimized: Written entirely in JAX, Waymax boasts in-graph compilation ensuring rapid simulations.
• Multi-Agent System: This feature enables a deeper understanding and modeling of intricate interactions among multiple road users, leading to fresh insights into multi-agent systems.

Researchers are granted easy access to the Waymax software via GitHub. Ready-to-use right after download, it includes the simplified representation of objects and bounding boxes from the Waymo Open Motion Dataset. This negates the necessity for users to employ their perception systems for raw sensor data interpretation. Additionally, Waymax comes equipped with pre-defined metrics and sim agents, simplifying the research trajectory.

Waymo's commitment to the research community remains unwavering. The tech giant is intent on incessantly refining the Waymax platform and eagerly anticipates the revolutionary outcomes it's bound to catalyze in the research landscape.

Source: Waymo