Skild AI Raises $300M to Develop “General Purpose Brain” for Plug-and-Play Robot Intelligence

Skild AI, a Pittsburgh-based robotics startup, has raised $300 million at a $1.5 billion valuation to develop a “general purpose brain” for robots that can be integrated across various applications.

Skild AI’s plug-and-play robotic intelligence: The company has created a foundational model that serves as a single off-the-shelf intelligence for robots, enabling them to perform basic functions:

• The AI model allows robots to navigate complex environments, such as climbing steep slopes, walking over obstructing objects, and identifying and picking up items.
• Skild AI’s model has been trained on a massive database of text, images, and video, which the company claims is 1000 times larger than those used by its competitors.
• The model demonstrates “emergent capabilities,” allowing robots to perform unanticipated tasks, similar to the behavior observed in advanced artificial systems like large language models.

Innovative data collection techniques: Skild AI’s cofounders, Abhinav Gupta and Deepak Pathak, have developed a mix of data collection techniques over years of research:

• Human contractors were hired to operate robots remotely and collect data about their actions.
• Robots were trained to carry out random tasks, record the results, and learn by trial and error.
• The AI model was also trained on millions of public videos.
• Pathak developed a way to instill “artificial curiosity” into robots, incentivizing them to explore more scenarios and collect more data.

Competitive landscape and industry reactions: Skild AI faces competition from several well-funded robotics companies that have emerged during the AI boom:

• OpenAI has recently revived its robotics team to supply models to robotics companies.
• Figure AI, a humanoid robotics company, and Covariant, an OpenAI spin-off building ChatGPT for robots, are among Skild AI’s competitors.
• Despite the competition, Skild AI’s access to large amounts of data is believed to set it apart from others in the space.

Envisioning the future of robotics: Skild AI’s founders aim to disrupt the robotics industry by enabling different use cases and products to be built on top of their foundational model:

• The company’s ultimate goal is to achieve artificial general intelligence for robots that people can interact with in the physical world.
• Investors believe that a “GPT-3 moment” is coming to the world of robotics, which will spark advancements similar to those seen in the realm of digital intelligence.

Analyzing the potential impact: Skild AI’s approach to creating a general-purpose robotic intelligence could revolutionize the industry, but challenges remain:

• While access to large amounts of data is crucial for scaling robotics, robots require specific types of data that may not be widely available on the internet.
• Translating data gathered from simulations to real-world applications can be difficult, and Skild AI will need to demonstrate the effectiveness of its combined data collection techniques.
• As the company works towards its goal of achieving artificial general intelligence for robots, it will need to address concerns surrounding safety, ethics, and the societal impact of advanced robotic systems.