Neo Humanoid Robot Learns New Skills with Advanced AI Training

Neo Robot's Enhanced Capabilities

• 🤖 Neo, a humanoid robot from 1X, is showcasing new capabilities through advanced AI training in employee homes.
• 💡 These new abilities include picking up unfamiliar objects, whole-body control, improved coordination, running, multidirectional walking, and navigating stairs.

Stair Navigation Without Lidar

• 🪜 Neo's new stair mode utilizes onboard cameras to estimate floor height, bypassing the need for Lidar sensors common in other robots.
• ⚠️ The robot demonstrated stepping sideways downstairs and recovering from a stumble.

Versatile Movement and Interaction

• 🤸 Neo can now perform tasks requiring it to get lower to the ground, such as sitting, kneeling, and standing up.
• 🛠️ Examples include handing over tools, working under a car, and crouching to load a washing machine.
• 🚶 The robot also exhibits pivoting, walking backward, and side-stepping for navigating tight spaces.

Redwood AI and World Model Updates

• 🧠 The new Redwood AI runs entirely on Neo's onboard GPU, enabling these advanced functions.
• 🌐 The 1X World Model bridges the gap between training simulations and real-world actions, trained on video, robot observations, and action trajectories.
• 🖼️ This model allows the robot to simulate future actions, with examples showing laundry, door opening, and cleaning tasks.

Learning from Experience and Failures

• 🍳 In an air fryer example, the AI initially hallucinated the basket was inseparable, but refined its simulation with gathered data.
• 📉 Like humans, 1X's robots learn from failures, with some examples shared by the company.
• ⚠️ The 1X World Model still faces challenges with simulating unfamiliar objects and predicting complex locomotion.

Future of General AI in Robotics

• 🎯 While a general AI model for robots is the ultimate goal, 1X acknowledges it will take time.
• 🌱 The company believes focusing on a variety of task-specific training goals will be crucial for progress.