Video Friday: Do Robots Even Need Legs?
Summary
This "Video Friday" brief from IEEE Spectrum Robotics highlights diverse advancements and applications. Genesis introduced Eno, an agentic robot integrating AI for real-world reasoning and action. NASA JPL field-tested ERNEST, a prototype rover, in the Colorado Desert, validating autonomous navigation software for future long-range lunar and Mars missions. Sony AI's Ace project demonstrated a table tennis robot simulating counterfactual ball trajectories for unpredictable conditions. Commercially, ANYbotics' ANYmal quadruped prevented a \$630,000 production loss in a concrete plant. Sanctuary AI achieved over 99.5% success and a 2.54-second cycle time for a complex automotive wire-plugging task. Other highlights include GITAI's satellite servicing preparations, Bi-AQUA's underwater robotics research, ABB Robotics' collaboration with PSYONIC on handling delicate objects using human-generated data, and GrowBot, a \$100 bipedal robot driven directly by an LLM on a Raspberry Pi Zero 2 W. The brief also lists upcoming robotics events for 2026.
Key takeaway
For AI Engineers developing autonomous systems, consider integrating agentic AI for complex real-world reasoning, as demonstrated by Genesis's Eno. If you are designing robots for unpredictable environments, explore real-time counterfactual simulation techniques like Sony AI's Ace project to enhance reliability. For robotics teams facing high-stakes industrial inspection, deploying quadruped robots can yield significant cost savings by proactively identifying critical infrastructure issues, as shown by ANYmal preventing a \$630,000 loss. Your next steps could involve evaluating these advanced AI and sensing integration strategies.
Key insights
Robotics innovation is rapidly expanding, integrating AI for autonomy, dexterity, and real-world problem-solving across diverse environments.
Principles
- AI agents enhance robot autonomy.
- Real-world data improves robotic dexterity.
- Simulation aids unpredictable environments.
Method
GrowBot employs an LLM to directly interpret raw IMU data and narrate its motion, leveraging a 50-Hz reinforcement-learning walk policy trained in simulation and transferred to the physical robot.
In practice
- Deploy quadruped robots for industrial inspection to prevent costly shutdowns.
- Utilize LLMs on low-cost hardware like Raspberry Pi for physical AI projects.
- Integrate human-generated data to train robots for delicate object handling.
Topics
- Agentic AI
- Autonomous Robotics
- Industrial Automation
- Robotic Dexterity
- Space Rovers
- LLM Robotics
Best for: CTO, VP of Engineering/Data, Director of AI/ML, Robotics Engineer, AI Engineer, AI Scientist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.