In the previous post we saw what we all knew was going to happen with humanoids, but they need to know where they are, where to go and what to avoid along the way and it is exactly that challenge that an American company is trying to solve. During the Nvidia GTC in California, Real Sense presented a new approach to making humanoids much more reliable in real-world environments.

The demonstration was carried out in collaboration with the Chinese company Limex Dynamics, showing a system capable of allowing robots to see, map and move precisely in three dimensions, all of this happening thanks to a combination of advanced 3D vision with something called cuVSLAM, a system that allows the robot to locate itself and map the environment at the same time.

It's like giving robots a kind of visual cortex, an ability to interpret the space around them in real time and that makes all the difference because unlike wheeled robots that operate on flat and predictable surfaces, humanoids need to deal with a much more complex world, stairs, inclines, obstacles, people moving. All of this requires constant decisions with a very high level of precision.

Traditional methods simply cannot meet these needs, systems such as basic sensors or 2D mapping cannot provide the necessary depth and spatial understanding, which is why many burglaries still depend on human control and fully controlled environments, but this new approach attempts to change that landscape.

With more advanced depth sensors and navigation systems, robots can identify changes in the terrain, avoid moving obstacles and adjust their steps in real time, they can climb stairs, avoid people and adapt to dynamic environments with much more safety and the most interesting thing is how they arrive at EBE, before entering the real world, they train a virtual environment. Using the Nvidia IS Club, the robbers go through extremely realistic simulations where they learn to balance, walk and make decisions.

It is as if they practiced thousands of times in a safe environment before taking the first step into the physical world. This process drastically reduces errors and speeds up development, and perhaps that is exactly what is allowing robots to stop being just demonstrations and begin to occupy real spaces in our physical world.

Only seeing the world is just half the challenge, because after identifying an object, an obstacle, even a person and robots, you need to interact with what you see and it is precisely at that point where most machines still fail, but that is changing.