Boston Dynamics has equipped its quadruped robot Spot with Google DeepMind’s Gemini Robotics-ER 1.6, aiming to push robots beyond scripted actions into reasoning-driven tasks.

The company showcased the upgrade in a video where Spot reads a handwritten to-do list and carries out tasks such as organizing shoes, picking up cans, and placing clothes in a laundry basket.

In one sequence, the robot takes a leash and walks a dog, highlighting its ability to translate natural language instructions into physical actions.

The system combines vision, language understanding, and task planning, allowing Spot to interpret its surroundings and respond with minimal human input.

This marks a shift from traditional robotics, where tasks required precise programming.

However, the demonstration also underscores the gap between AI reasoning and real-world execution. In one example, the robot grips a can sideways, a simple mistake that could lead to spills, showing that human-like understanding remains limited.

Inspection drives real value

Despite the home-focused demo, the primary goal of the upgrade is industrial inspection, where Spot is already deployed at scale.

The robot is designed to navigate facilities, identify hazards, and monitor environments where human access may be difficult or risky.

With Gemini Robotics-ER 1.6, Spot can autonomously detect issues such as pooled water, read gauges, and interpret site conditions. It can also call on vision-language-action models to better understand complex environments.

“Advances like Gemini Robotics ER 1.6 mark an important step toward robots that can better understand and operate in the physical world,” said Marco da Silva, Vice President and General Manager of Spot at Boston Dynamics.

“Capabilities like instrument reading and more reliable task reasoning will enable Spot to see, understand, and react to real-world challenges completely autonomously.”

The update builds on a partnership between Boston Dynamics and Google DeepMind announced earlier this year, focused on integrating advanced AI models into robotic systems.

Limits of robot reasoning

While the system improves usability, challenges remain. The model currently relies heavily on vision-based data and lacks the deeper physical understanding that humans gain through touch and experience.

“For robots to reliably and safely perform tasks, this connection between how robots understand the world and how humans do is critical,” said Carolina Parada, Head of Robotics at Google DeepMind.

The company is working to improve safety and reasoning through benchmarks that guide how robots should behave in real-world scenarios.

However, full autonomy still depends on achieving consistent reliability in varied environments.

Boston Dynamics says real-world deployment will continue through controlled rollouts, with customer feedback playing a key role in refining the system.

The company notes that robots must reach a reliability threshold high enough to avoid false alerts and maintain user trust.

The development highlights both the progress and limitations of embodied AI, where intelligence is integrated into physical systems.

While Spot’s ability to walk a dog may capture attention, its real-world impact lies in improving safety and efficiency in industrial operations.