VERSES describes multi-agent robotics using hierarchical active inference. Source: VERSES AI
Part of the challenge of combining artificial intelligence with robots is the amount of training data required. VERSES AI Inc. yesterday said its robotics architecture can accomplish typical household tasks better than other robotics models and without any pre-training.
“I believe that by combining our world modeling and our active inference capabilities, we’ve shown robots can think on their ‘feet’ — navigating and completing complex tasks without months of costly training.” stated Hari Thiruvengada, chief technology officer of VERSES. “Our breakthrough has the potential to transform how robots operate across industries, from factories and warehouses to homes and public spaces, potentially unlocking a new era of truly adaptive, reliable automation.”
Founded in 2020 as Verses Technologies Inc., VERSES said it is “a cognitive computing company building next-generation agentic software systems” inspired by nature. The Vancouver, British Columbia-based company claimed that it designed its Genius flagship product around first principles found in science, physics and biology. The platform can generate reliable domain-specific predictions and decisions under uncertainty, according to VERSES.
In its second-quarter 2025 filing this week with the U.S. Securities and Exchange Commission, VERSES reported net revenue of $115,939, a net comprehensive loss of $9.4 million, and $3.2 million in cash.
Editor’s note: RoboBusiness 2025, which will be on Oct. 15 and 16 in Santa Clara, Calif., will include the Physical AI Forum among other tracks. Register now to attend!
VERSES builds AI for adaptability
VERSES described robots as falling into two categories: drive-by-wire, in which everything is pre-programmed, and deep learning, which requires large amounts of data for training.
“Robots often perform well on scripted tasks but can freeze when faced with new situations; even something as simple as a box in the wrong place can halt progress,” noted the company, which cited autonomous guided vehicles (AGVs) as examples of drive-by-wire systems.
Since factories, warehouses, and homes are always changing, robots often struggle to adapt, working more slowly or even stopping, it said. To overcome their inherent limitations, robotics environments are often controlled. For instance, robots may be placed in a cage or in areas where no humans are allowed. This practice greatly reduces the robots’ usefulness, said VERSES.
While deep learning approaches are more flexible, they require a lot of data and can still struggle with changes, such as a bottle falling over or a chair being out of place, according to the company.
“When a human needs to get a drink in a new apartment, they don’t execute by having practiced this task in hundreds of different apartments; they are able to adapt because they have a model of how the world works,” said VERSES. “This allows humans to figure out that they need to open the refrigerator and grab a bottle.”
The company asserted that its system doesn’t require any pre-training and instead just adapt by exploring the environment. They consist of vision, planning, and control modules, enabling robots to handle unexpected obstacles or to pick up dropped items.
New models tackle household tasks
Members of VERSES’ research lab have published a paper entitled, “Mobile Manipulation with Active Inference for Long-Horizon Rearrangement Tasks.” They compared the VERSES robotics model is with a deep learning alternative in three tasks: tidying a room, preparing groceries, and setting a table.
VERSES says its model performed well against the baseline in basic household tasks. (Click here to enlarge.) Source: VERSES AI
The VERSES robotics model achieved a success rate of 66.5% across these tasks, while the previous best alternative had a success rate of 54.7%. The company claimed that its model required no training, whereas the “multi-skill mobile manipulation” model required 1.3 billion steps to pre-train several skills across the three tasks.
The VERSES model had basic knowledge such as the resting pose of its own arm when idle or how much resistance the arm would get from obstacles. By contrast, the baseline model required offline training of 6,400 episodes per task and 100 million steps per skill across a total of seven skills, such as picking up an object or opening a fridge.
“Currently, robotics systems are often brittle and need huge amounts of training data, which makes them expensive and prone to going wrong.” said Sean Wallingford, former president and CEO of Swisslog, a leading logistics automation company. “For instance, if you bring a robot to a new factory or ask it to do a different job, it will need a lot of re-training and may not be reliable.”
“VERSES breakthroughs are exciting, because they offer an alternative approach,” he added. “If we can deploy robots without training, they will be viable in a wide range of activities, from factories and warehouses to domestic and commercial applications.”
