The Rise of Physical AI Agents

Not very long ago, NVIDIA chief Jensen Huang said that the next wave of AI will be physical. He believes that robots powered by AI will become part of every industry and revolutionise them. Surprisingly, the era of physical AI, or embodied AI, has already begun, and last week witnessed a slew of announcements. (see below)

Giving artificial intelligence a body: Figure Robotics’ ubercool newest humanoid robot, Figure 02, was revealed a few days ago. The company’s founder and CEO, Brett Adcock, released a demo video that showcased the robot’s dexterity and movements, predominantly showcasing its capabilities in a trial run at the BMW Group Plant Spartanburg.?

Arm wrestling, anyone? Figure 02’s 16 degrees of hand dexterity is a major breakthrough, especially given the complexity of humanoid robotics. Adcock noted that while five-fingered hands are crucial for general robotics, their complexity often deters traditional roboticists. With little progress beyond prosthetics, the Figure team had to design these hands from scratch.?

OpenAI, Microsoft, and NVIDIA have been pivotal in powering both the hardware and software for Figure. Figure 02 leverages NVIDIA’s Isaac Sim for synthetic data and AI model training, with NVIDIA enhancing simulation, training, and inference through its full-stack systems. Microsoft has contributed H100 NDs for large model training, while OpenAI has fine-tuned custom multimodal models using humanoid robot data.

Where is Tesla’s Optimus? Interestingly, there has been no update from Tesla since they released the Optimus Gen 2 in December last year.?

Funnily, Adcock took a dig at Elon Musk after Figure 02’s launch as a challenge to already existing humanoids in the market. At the same time, demo videos comparing both the Optimus Gen 2 and Figure 02 were doing the rounds, with the latter proving to be similar to the Tesla humanoid.

Meanwhile, Google DeepMind is busy training its robot to play ping-pong. It has managed to achieve a 45% win rate against human opponents by adapting its tactics and refining its skills in real-time, marking a significant step towards robots performing complex tasks in real-world environments.

Chinese AI startup Guoxing Intelligent has developed firefighting robots equipped with thermal cameras and smoke detectors, capable of remotely assessing fire scenes, extinguishing flames, and locating trapped individuals, all while providing real-time communication to enhance rescue operations.

In India, Bengaluru-based Control One is pioneering the next Boston Dynamics with its vision-based AI agents, transforming forklifts into autonomous, voice-controlled machines.

Leveraging Visual SLAM instead of traditional LiDAR, Control One’s ‘One AI’ operating system enables forklifts to navigate and execute tasks intelligently.?

Despite minimal funding, the startup has attracted significant investment and aims to address the global labour shortage in warehouses by enabling remote management of entire robot fleets. Their focus on efficient, on-edge computing with NVIDIA GPUs sets a new standard for intelligent material handling.

In focus: Backed by Khosla Ventures, Lockheed Martin, and Remy, AI and robotics startup Vayu Robotics, started by Indian founders Mahesh K, Anand Gopalan, and Nitish Srivastava in 2022, has been making the rounds for their work in developing physical AI agents that are low-cost, environmentally sustainable and scalable.

Breakthroughs: MIT researchers recently came up with a new EES (Estimate, Extrapolate, and Situate) algorithm that enables robots to autonomously adapt and excel in new environments by efficiently practising and refining skills like sweeping and object placement.

Impact: Grain Weevil, an innovation by Nebraska-based Zane Zents and Chad Johnson, is revolutionising agricultural safety with a cutting-edge robot that automates grain management, keeping farmers out of hazardous grain bins and significantly reducing the risk of injuries and fatalities.

Fashion, Beauty & Lifestyle: LUUM Lash, founded by Nathan Harding and Jo Lawson, is revolutionising the beauty industry with the first AI-powered robot for automated eyelash extensions, promising faster, safer, and more precise applications, and potentially transforming beauty services globally.?

Christina Ernst, a Google software engineer, took the internet by storm with her generative AI-powered dress, featuring robotic snakes that detect and follow faces, alongside inspiring young girls in robotics through her platform, She Builds Robots.

Education: Another friendly robot called Moxie, created by Paolo Pirjanian, founder of Embodied Inc., is a multimodal AI robot designed to support children's therapy, education, and emotional development, with plans to enter the Indian market next year.

Three-body problem: Huang recently said that the future of physical AI depends on three distinct computing systems — “one to create AI, another to simulate AI using synthetic data, and a third to actually run the AI.”?

This “three-computer problem” is incredibly complex, akin to a “three-body problem” in physics. It resonated with Huang, who said that NVIDIA has pioneered this approach, developing an entire stack for robotics, from the computing infrastructure to the AI models and safety systems.

For instance, NVIDIA's Project GR00T uses real and synthetic data to train humanoid robots, leveraging the MimicGen NIM to expand datasets and refine models in both simulation and real-world tests. The NVIDIA OSMO Robotics Cloud orchestrates this process, enabling developers to push the boundaries of AI-powered robotics.

AI Events >>

Cypher 2024?

India’s biggest AI conference is just around the corner. Mark your calendars for Cypher 20204! The Cypher India edition is scheduled to be held in Namma Bengaluru from September 25 to 27, while Cypher US will be held in Santa Clara, California from November 21 to 22, 2024.?

What are you waiting for? Register today!?

Cypher 2024 | India | USA?

AI Outlook >>?

100 Most Influential AI Leaders in India 2024

AIM has curated its 2024 list of the 100 Most Influential AI Leaders in India, celebrating those who are driving innovation, overcoming challenges, and setting new benchmarks in the AI landscape, alongside shaping the future of technology in the country.

Checkout the full list to see the leaders shaping India's AI future!

AI Insights >>?

GenAI Startup Landscape in India 2024?

AIM Research recently released a first-of-its-kind report called the GenAI Startup Landscape in India 2024, highlighting the rapid growth of generative AI startups in India.

Key Insights:

• The global GenAI market is projected to grow from $39.2 billion in 2023 to $1082.4 billion by 2033, with a CAGR of 40.22%.
• India hosts over 150 GenAI startups, with 51.4% concentrated in Namma Bengaluru.
• Emerging tech hubs like Hyderabad contribute 7.2% of GenAI startups, driven by strong IT and pharmaceutical sectors.
• Chennai and Ahmedabad are also growing as tech hubs, with Chennai focusing on SaaS and Ahmedabad benefitting from institutions like IIM Ahmedabad.

Read the complete report here.?

AI Nuggets >>?

• Google, NVIDIA, and other major companies recently proposed investments worth INR 3,200 crore in Madhya Pradesh during a Bengaluru event, aiming to boost the state’s IT sector, create 7,000 jobs, and establish it as a key investment destination.
• Amgen is launching a biotechnology and AI hub in Hyderabad’s HITEC City, which is expected to be operational by Q4 2024 and will enhance digital capabilities across its global operations, creating up to 3,000 jobs in the AI, data science, and life sciences fields.
• Hugging Face recently acquired XetHub to enhance its platform’s capabilities in managing large AI datasets and models, aiming to streamline AI development and versioning as models grow in size.
• Lentra, a Pune-based AI-driven SaaS lending platform, has joined AWS’ ISV Accelerate Programme, enhancing its ability to deliver digital lending solutions to major banks like HDFC and Citi, while leveraging AWS’ cloud computing to drive innovation and streamline services.
• Mistral recently introduced the La Plateforme Agent Builder for non-technical users, alongside the Agents API for developers, enabling easy integration of agent creation into existing workflows and offering extensive customisation options.
• Genpact appointed Sanjeev Vohra as its first chief technology and innovation officer to drive AI initiatives, alongside accelerating the company’s digital transformation and innovation strategy.