355
Audio & Video Production344
Automation & Workflow224
Software Development250
Marketing & Growth192
AI Infrastructure & MLOps174
Writing & Content Creation203
Data & Analytics140
Design & Creative169
Customer Support131
Photography & Imaging156
Sales & Outreach125
Voice & Speech135
Education & Learning131
Operations & Admin87
Decart released Oasis 3, a tool that generates realistic driving environments in real time for testing self-driving systems, with access offered through an API.
In short: Decart has released Oasis 3, a system that can generate realistic driving environments in real time, and it is available to developers through an API.
AI startup Decart unveiled Oasis 3, which it describes as an interactive “world model” for driving. A world model is like a video game world that an AI creates as you go, frame by frame (like drawing the next page of a flipbook based on the last page).
Decart says Oasis 3 can create photorealistic, multi-camera driving scenes for testing self-driving technology. It can show a front view and side views, and it can keep generating new road scenes for long periods. The company is selling access through an API, which is a way for other software to “plug in” to a service, like connecting an app to a payment provider.
Pricing starts at $0.02 per second of use, and Decart says larger customers will have custom pricing. The company is initially aiming at autonomous vehicle companies, and it says it plans to expand to robotics and other uses later.
TechCrunch’s testing found strong, realistic scenes at the start, but the experience could break down over time. Places could drift from the original setting, earlier intersections could disappear when returning, controls could feel unresponsive, and cars could sometimes pass through other cars, which shows limits in how well the system follows real-world physics.
Self-driving car testing needs rare and dangerous situations that are hard to stage in the real world. Tools like Oasis 3 can help companies practice those “edge cases” in a safer, repeatable way, although the current gaps mean it may not fully replace other kinds of testing yet.
Source: TechCrunch AI