Vention releases Rapid Operator AI to automate deep bin picking

Vention Inc. has developed Rapid Operator AI to automate complex, unstructured tasks, beginning with deep bin picking. The company announced the system’s commercial launch at NVIDIA GTC 2026 last week.

“Rapid Operator AI is a productized, physical AI solution for unstructured manufacturing tasks. I’m not talking about warehousing here; I’m talking about manufacturing,” Etienne Lacroix, the founder and CEO of Vention, told The Robot Report. “The world of manufacturing is significantly more demanding.”

Lacroix said the new product is built on the company‘s Generalized Robotic Industrial Intelligence Pipeline (GRIIP). GRIIP delivers a unified pipeline from perception to motion by integrating Vention’s proprietary models with NVIDIA Isaac open models.

Vention is targeting midmarket and enterprise manufacturers operating multi-shift facilities where labor shortages and high production variability create operational strain with the system.

Why start with deep bin picking?

Vention highlighted two reasons for targeting deep bin-picking tasks. First, its customers said it was a common problem.

“When we talk to customers in the industry, it’s just a very recurrent problem. In assembly or machine tending, you have a bin of parts, and then you have to take them out of the bin and then do an operation with them,” explained Francois Giguere, chief technology officer at Vention. “So, it’s a use case that very often has blocked us, because we didn’t have a scalable way to adapt to this type of environment.”

“Now, leveraging these new technologies, we’re in a much better position to say yes to these projects and implement something for the customers,” he added. “Everything comes in these big, deep bins. They have a fixed form factor, and they’re part of their operation, so you have to deal with it.”

The second reason Vention started with bin picking was because of how challenging the task was. Picking deeply in bins adds a lot of complexity, It’s hard to see what you’re trying to pick, and you need to ensure the robot or camera doesn’t collide with the bin itself or objects within the bin, Lacroix said.

However, the team knew that if they could tackle this issue, they would be able to tackle any other one in manufacturing.

“The first deployment we did was a client that had four different attempts to solve this with traditional vision,” recalled Lacroix. “Each of them had failed to the point that when we proposed to them this kind of use case as an R&D case for us to bring this technology to market, they were skeptical.”

Vention on building an efficient and flexible AI model

Vention said Rapid Operator allows robots to:

• Detect randomly oriented parts in dense clutter, estimate precise 6-DoF (degree-of-freedom) pose, and plan collision-free grasps
• Execute autonomous picks with adaptive retries for reliable, multi-shift operation with minimal supervision
• Support opaque, translucent, and transparent materials; perform in bright light, low light, or darkness; handle containers up to 24 in. (60.9 cm) deep

To make a system that can do all of this quickly, Vention needed to take the best parts of AI pipelines and world models.

“AI pipelines are super efficient. They’re fast, they’re able to meet industrial-grade cycle times. World models, like the ones we very often see these days on humanoids, are very generalizable, but they’re slow and cannot meet the usual cycle times of manufacturers,” said Lacroix. “So, how do you get the best of both? You want generalization, and you want speed and performance.”

NVIDIA plays a role in development

Vention uses NVIDIA FoundationStereo for stereo matching, and NVIDIA FoundationPose for pose estimation.

“Building foundation models from scratch requires a lot of compute. It’s extremely expensive. Building these models also requires a lot of expertise,” Giguere said. “So, we’ve let [NVIDIA] do that portion of the effort, and we’ve integrated that into a pipeline for applications.”

Looking ahead, Lacroix said Rapid Operator AI will remain a manufacturing-focused system. However, with GRIIP, the company can offer a wider variety of tasks.

“Any manufacturer that operates a two-shift factory can now deploy physical AI within a two-year payback,” Lacroix said. “You get the speed of humans, the reliability of humans in terms of pick, and you’re able to navigate, at the same time, those very intricate, very constrained manufacturing environments without any collision.”

---

---