Vayu Drive is a machine learning model with a sim-first approach that allows for representation learning, robust behavior planning and a fast iteration cycle. With Vayu Drive, customer deployment is faster, easier and more efficient than ever. By reducing dependence on expensive hardware and services as well as real-world data collection and annotation, Vayu Drive creates a network effect enabling deployment across multiple domains.
Vayu Drive is a transferable autonomy stack that can adapt across different domains. By using a sim-first approach, Vayu Drive allows for representation learning, robust behavior planning and a fast iteration cycle. Only 100 hours of real-world data collection enables Vayu Drive to work across multiple operational domains, making it an ideal solution for a diverse set of industries and applications.
An end-to-end learned drive agent designed to work on a single embedded SOC with minimal latency
Fully end-to-end learning
Deep learning-based planner makes the robot resilient to uncertainty from new, complex or confusing situations.
A “vision-based” navigation approach (a.k.a “online mapping”) means there is no reliance on lidar-based or HD mapping to achieve autonomy.
Automated hard scenario generation for training the planner
A scalable validation and verification approach. We validate in sim against adversarial agents, which leads to a fast cycle of improvement.
