Boston Dynamics' Robots Can Interpret Gauges and Thermometers with Nearly Perfect Accuracy

Advancements in the field of robotics and artificial intelligence are profoundly transforming the way machines interact with the real world. A significant example comes from the collaboration between Boston Dynamics and Google DeepMind, which has led to a major upgrade in the capabilities of the quadrupedal robot Spot.

Thanks to the new model Gemini Robotics-ER 1.6, Spot is now able to read analog instruments such as thermometers and pressure gauges with high precision during its inspections in industrial environments. This represents a remarkable step forward, as such tasks require complex visual interpretation, including the recognition of hands, liquid levels, numerical indicators, and other variables present in traditional instruments.

The new model introduces the concept of "agentic vision", a capability that combines visual reasoning with the execution of code-like operations to analyze images. This system allows the robot to create a kind of "digital visual board" on which to process the collected information. The result is an impressive performance improvement: the accuracy in reading the instruments jumps from 23% in the previous model to around 98%. Even without this advanced feature, the system still achieves high levels of accuracy.

Further Innovations Introduced by Boston Dynamics

Another innovative element is "multi-view reasoning", which allows the robot to use multiple video streams simultaneously to better understand the surrounding environment. This makes Spot particularly suitable for operating in complex contexts like factories, warehouses, and industrial plants, including those of the Hyundai Motor Group, the owner of Boston Dynamics.

In addition to precision, Google DeepMind has also paid great attention to safety. The new model is designed to adhere to physical constraints and assess risks, such as the presence of humans or potentially dangerous situations. For instance, it can recognize scenarios where a person might be exposed to hazards, thus improving the interaction between robots and human operators.

Despite these advancements, robots have not yet reached a level of understanding comparable to that of humans. However, technologies like Gemini Robotics-ER 1.6 represent an important step toward a future where robots can operate more autonomously and flexibly, moving beyond the repetitive tasks of production lines to become true intelligent assistants in real-world environments.