EP.37 A ROBOT DOG THAT SKATES
EV maker employs humanoids, synthetic robot data, AI that generated 3D models & much more ...
🚨 BREAKING:
Newsletter subscribers have earlier!
I am launching my Robocommunity on Discord - 'The Robosapiens'.
There will be an opportunity to get to know each other better, make new friends, and exchange information and insights on robotics!
The server is new, and only newsletter subscribers have access!
May The Robot Be With You
This robot dog just got the skates! 🛼
Unitree Robotics has unveiled its latest innovation in quadruped robotics - the Go2-W, a wheeled variation of its popular low-cost Go2 platform. This new model replaces the traditional "paws" with wheels, creating a unique kinematic design that enhances the robot's agility and efficiency on hard surfaces while enabling new capabilities.
What sets the Go2-W apart is its locomotion versatility. The robot can roll on two to four wheels as needed, providing smooth and efficient movement on flat surfaces. In this mode, the legs act as active shock absorbers, ensuring a stable ride for the robot's body and any onboard sensors. When faced with obstacles, the Go2-W can lock its wheels and switch to a traditional quadruped walking mode, capable of climbing over barriers up to 70 cm in height.
This mixed-modality approach isn't entirely new in the quadruped world. Companies like LimX have previously demonstrated similar concepts. The idea behind wheeled quadrupeds is to combine the efficiency of wheeled locomotion on smooth surfaces with the versatility of legged movement for navigating complex terrains or obstacles like stairs.
While Unitree hasn't disclosed the pricing for the Go2-W, it's worth noting that the entry-level Go2 starts at $1600 USD. This suggests that the Go2-W could potentially offer an affordable entry point into the world of mixed-modality robotics, continuing Unitree's trend of providing accessible quadruped platforms.
Robot data with synthetic simulations! 👾
Researchers at NVIDIA's GEAR Lab have made a breakthrough in scaling up robot data collection with Project GR00T. This innovative approach tackles one of the most significant challenges in robotics by multiplying human-collected demonstration data in simulations, potentially increasing the dataset by 1000 times or more.
The project utilizes a three-step process to achieve this remarkable data expansion. First, Project GR00T employs Apple Vision Pro to provide human operators with first-person control of humanoid robots. This advanced system parses human hand poses and retargets the motion to the robot's hand in real-time, creating an immersive "Avatar-like" experience for the operator. While teleoperation is typically slow and time-consuming, this method efficiently collects a small but crucial initial dataset.
Next, NVIDIA's generative simulation framework, RoboCasa, is used to multiply the demonstration data by varying the visual appearance and layout of the environment. For example, a single cup-placing demonstration in one physical kitchen can be replicated across hundreds of virtual kitchens with diverse textures, furniture, and object placements.
The final step involves applying MimicGen, which multiplies the data by varying the robot's motion. MimicGen generates numerous new action trajectories based on the original human data and filters out unsuccessful attempts to create a much larger, high-quality dataset.
This synthetic data pipeline effectively breaks the barrier of physical limitations in robot data collection. By trading compute power for expensive human data through GPU-accelerated simulations, Project GR00T opens up new possibilities for scaling in robotics, similar to the advancements seen in Large Language Models (LLMs).
AI model that generates 3D assets! 🧩
Stability AI, the London-based startup behind the image generation model Stable Diffusion, is still fundraising to close its $80m round. Despite reports that the company had secured the full amount, sources reveal that Stability AI has only raised around $50m so far and is seeking an additional $30m from existing investors.
Stability AI's fundraising is notable for its scale, particularly compared to other recent raises by GenAI startups. Paris-based Mistral raised €600m last month, while German Aleph Alpha raised €500m in November 2023. As the demand for AI-powered image generation continues to grow, Stability AI is well-positioned to capitalize on this trend.
The additional funding will likely be used to drive the development of Stable Diffusion and expand the company's operations. With its new leadership in place, Stability AI is entering a new era of growth and innovation.
Meme of the week 🤖
My first meme - so please do not judge. This meme became a classic during Olimipada. This is my interpretation. At a time when we have new updates about humanoids every day, and new applications of real use cases are non-existent.
Drone log analysis using MATLAB! 🛸
In this course, participants will learn the intricacies of flying drones, both manually and autonomously, understanding the critical components such as chassis, motors, and sensors, and identifying potential reasons for crashes. The focus is on tuning drone performance through simulation and post-flight analysis.
Upon completion, learners will be able to:
Utilize the Flight Log Analyzer app to analyze UAV autopilot flight logs.
Import custom flight logs and signal data for detailed analysis.
Identify anomalies in-flight data to troubleshoot and improve drone performance.
Humanoids work for Zeekr now! 🔋
Zeekr, the premium electric vehicle (EV), is piloting the use of humanoid robots in its factory. This innovative move follows in the footsteps of Nio, another EV maker that has already experimented with humanoid robots in its production lines.
The humanoid robot, supplied by UBTech, has been working at a Zeekr factory for 21 consecutive days, performing tasks such as lifting and carrying. This collaboration between UBTech, Geely Holding, and Miracle Automation Engineering aims to drive the scaling of humanoid robots in automotive and component smart manufacturing scenarios.
The Walker S Lite humanoid robot has been designed to work seamlessly with automated guided vehicles (AGVs) and navigate different assembly lines with ease. Equipped with advanced technologies like semantic Visual Simultaneous Localization and Mapping (VSLAM) navigation and whole-body fine motion control, the robot can walk stably while lifting 15-kilogram boxes.
Zeekr is not the only EV maker to explore the potential of humanoid robots. Nio has already experimented with UBTech's Walker S robot in its factory, utilizing its real-time image capture and transmission capabilities for quality inspections and affixing vehicle logos.
As the demand for EVs continues to grow, the use of humanoid robots in manufacturing is likely to become more widespread. With their ability to work alongside humans and automate complex tasks, humanoid robots are set to play a key role in shaping the future of the EV industry.
Keep reading with a 7-day free trial
Subscribe to we all are robots to keep reading this post and get 7 days of free access to the full post archives.