Singapore start-up Sapiens Technology says it has cut the training time for robotics AI from weeks or months to as little as 30 minutes. If the company’s account holds up, that could make it quicker to adapt robots for different tasks, rather than retraining an entire model each time.
Watch Desk analysis
What happened
Sapiens says its method trains only 1 billion to 3 billion model parameters, focusing on elements it says can apply across scenarios, such as understanding an operating environment. Training for a particular task then changes the testing environment, tasks and evaluation. The Straits Times reports that the company demonstrated the approach with commercial robot arms and off-the-shelf robot ducks picking up objects. The reported training times range from 30 minutes to less than a week, compared with weeks to months typically required for a robotics application.
Sapiens says trained robots can adapt to changes in tasks and respond to text prompts for variations, without detailed instructions being coded for each one. The company also says it can make a robot’s reasoning readable for human review and add a delay before planned actions are executed, allowing a person or monitoring system to stop them.
The company is in talks with more than 30 robotics and medical firms. TetraBOT, which has tested the method, is planning to train more industrial robots; Shanghai Xinwei Medical Technology is working with Sapiens on inventory forecasting and business insights, according to the report.
Why it matters
Faster training could make custom robotics more practical for companies whose tasks or operating conditions do not fit off-the-shelf systems. Sapiens points to industrial robots operating switches and controls in electrical substations, where workers may otherwise need protective equipment, and to medical supply forecasting as possible applications.
The reported demonstrations and customer work give this claim more shape than a speed figure alone. But the training times, adaptability and auditability are company claims reported by The Straits Times, not independent benchmarks. The difference between a promising method and a dependable robot is still measured in real-world performance, not a brisk stopwatch.
Our read
This is a worthwhile robotics development because it addresses a practical bottleneck: adapting AI systems to specific machines and jobs. The most intriguing part is the claimed combination of quicker task training and a readable, interruptible action process. The next proof should come from results beyond demonstrations and from customers using the method in demanding settings.
What to watch
- Whether TetraBOT moves from testing to wider deployment, and what the robots can reliably do.
- Whether Sapiens publishes independent comparisons of training time and performance.
- How the claimed monitoring and stop mechanisms work during real operations.
Discussion spark: Would you trust a robot trained this quickly in a high-risk workplace if its reasoning is readable and its actions can be interrupted, or should longer testing be a hard requirement?
Sources and evidence
- Source update (11 November 2026, 07:23 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.