Remember the day when we used to see Robots on TV or in the dream world? Yes, those days have now come in the real world. A robot is a machine, especially one programmable by a computer, which is capable of carrying out a complex series of actions automatically. Today, we are in a world where AI (Artificial intelligence) agents have taken over the robot. Qualcomm is now working on generative modelling applications to embodied AI and robotics.
Since the robots have a narrow goal and can only learn from their surrounding atmosphere a little, Qualcomm started working to stretch out the limitations of robots through embodied AI. For more information, these agents are furnished with sensors (for instance, cameras, pressure sensors, accelerometers, and many more.) that catch data from their surrounding environment, along with AI systems that can easily analyse and even learn from the obtained data.
What Qualcomm Is Doing For Embodied AI
Qualcomm is working on applications of generative modelling to embodied AI and robotics, aiming beyond classical robotics and empowering with the capabilities as per the convenience of modernity and requirements. The brand is especially trying hard to offer capabilities which are mentioned below;
- Open vocabulary scene understanding.
- Natural language interface
- Reasoning and common sense via large language models (LLMs)
- Closed-loop control, dynamic actions via LLMs or diffusion models
- Vision-language-action models.
Somewhere, Robotic machines have a drawback of all these mentioned capabilities. They need data efficiency, low latency, enhanced privacy and sensor processing. Fortunately, all these convenient facilities can be achieved via on-device AI.
It is the only reason Qualcomm Technologies is working to develop platforms to support the creation of more advanced and productive robots, for instance, the Qualcomm Robotics Platform. These platforms comprise the Qualcomm AI Engine, offering capabilities that can lose original applications and possibilities.
Qualcomm is also targeting the drawbacks by which they get solved. In some scenarios, internet AI learns from static datasets to resolve various queries and tasks, but sometimes, related to that query or task, data is not readily available on the internet and is expensive to acquire. To resolve such troubles, the Qualcomm AI Research team has introduced a novel data-efficient architecture model to increase the perception of robots in their environment.
The brand named this architecture Geometric Algebra Transformers (GATr). It has the expandability and expressivity of transformers. It is capable of showing a performance even after a tiny amount of data. It is a general-purpose mechanism for geometric data. The brand provides three countable components: Geometric algebra representations, equivariant layers and a transformer architecture.
In short, the brand is working to provide an efficient understanding of 3D images or videos with AI, capture data from their surroundings, and many more capabilities in On-device generative AI to play a fundamental role in embodied AI.
“If you like this article follow us on Google News, Facebook, Telegram, and Twitter. We will keep bringing you such articles.”
