The widespread adoption of industrial robots has long been hampered by the immense challenge of training them for diverse, real-world tasks. Unlike large language models that benefit from vast online datasets, embodied AI for robotics faces a severe shortage of relevant training data. Reimagine Robotics, a London and Sydney-based startup, is tackling this problem by empowering human workers to become 'robot teachers,' enabling on-the-job learning without the need for specialist programmers.

Led by CEO Jonathan Scholz, who previously headed DeepMind's applied robotics team, Reimagine Robotics employs a "monkey see, monkey do" methodology. Workers can demonstrate a task, observe the robot's attempt, and then physically guide or correct the robotic arm. This interactive process allows for rapid iteration and adaptation, significantly reducing the time required to teach a robot new behaviors. In one manufacturing deployment, the startup reportedly cut the training time for a specific task from one day to just 10 minutes.

This approach marks a departure from traditional, code-intensive robot programming, which often requires highly specialized expertise and is time-consuming. By democratizing the training process, Reimagine Robotics aims to make advanced automation accessible to a broader range of businesses, particularly in manufacturing and electronics disassembly. The company has already deployed its robots in various facilities, including a plastics business where workers trained robots to tend 3D printers and automate further production stages.

Reimagine Robotics, which emerged from stealth earlier this month and is backed by Fly Ventures and Firstminute Capital, is not just simplifying robot deployment; it's fostering a new paradigm where human-robot collaboration is central to operational efficiency. This model suggests a future where robots are integrated as adaptable tools, constantly learning and improving alongside their human counterparts, rather than static, pre-programmed machines.