I caught Pras Velagapudi in the hallway after a breakout session at the Humanoid Summit. I promised it would take less than two minutes. We finished in one.

Velagapudi is the CTO of Agility Robotics, the company behind Digit, one of the few humanoid robots already working in real-world commercial environments. I asked him two straightforward questions. What’s new and what can we expect over the next year?

Quite a lot, it turns out.

Digit has now been deployed in additional locations, including a newly signed contract with Mercado Libre. That matters because it reinforces something important. Digit is not a demo robot. It is operating in logistics and manufacturing environments today.

Those deployments are running on Agility’s current V4 platform. But the bigger story is what comes next.

Agility is actively working on its next-generation system, the V5 platform, planned for release next year. According to Velagapudi, V5 is designed to support more use cases and, crucially, to incorporate what he described as onboard cooperative safety.

This is where the deep learning story really begins.

Cooperative safety means Digit can operate in the same physical spaces as people while performing different tasks, without requiring strict separation or constant handoffs between humans and robots. That capability is not just a hardware problem. It is fundamentally an AI problem.

For a robot to share space with people safely, it needs to perceive human motion, understand intent at a basic level, adapt its behavior in real time, and recover gracefully when the environment changes. That requires a stack of learned behaviors layered on top of classical control systems.

During his presentation at RoboBusiness, Velagapudi addressed the same concern talking about what needs to be done before humanoids like Digit can work safely in the same space as people. Check out clips from panel discussion here.

Velagapudi also pointed to what Agility expects over the next twelve months. We should start seeing sneak peeks of the V5 platform and new capabilities emerging from Digit’s AI-powered skill stack.

That phrase is doing a lot of work.

A skill stack implies modular, learned behaviors that can be composed, updated, and extended. It suggests a shift away from hard-coded task execution toward systems that can generalize across tasks and environments. In other words, this is not just about better walking or lifting. It is about embodied intelligence.

This short hallway conversation reinforced something I have been hearing repeatedly across the robotics industry. The next wave of progress is not coming from flashier hardware alone. It is coming from tighter integration between perception, learning, and control.

Digit’s evolution from V4 to V5 is a good example of that shift in motion. I first saw Digit at NVIDIA GTC back in March, where it was operating on the V4 platform. It’s exciting to now see how that foundation is evolving.

As I like to call this clip from March, it is Digit having fun on its’ Target run.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit dianawolftorres.substack.com