The ALOHA Unleashed method builds on our ALOHA 2 platform that was based on the unique ALOHA (a low-cost open-source hardware system for bimanual teleoperation) from Stanford University.
ALOHA 2 is significantly more dexterous than prior systems since it has two hands that might be easily teleoperated for training and data collection purposes, and it allows robots to learn how you can perform recent tasks with fewer demonstrations.
We’ve also improved upon the robotic hardware’s ergonomics and enhanced the educational process in our latest system. First, we collected demonstration data by remotely operating the robot’s behavior, performing difficult tasks like tying shoelaces and hanging t-shirts. Next, we applied a diffusion method, predicting robot actions from random noise, just like how our Imagen model generates images. This helps the robot learn from the information, so it may well perform the identical tasks by itself.
Learning robotic behaviors from few simulated demonstrations
Controlling a dexterous, robotic hand is a fancy task, which becomes much more complex with every additional finger, joint and sensor. In one other recent paper, we present DemoStart, which uses a reinforcement learning algorithm to assist robots acquire dexterous behaviors in simulation. These learned behaviors are especially useful for complex embodiments, like multi-fingered hands.
DemoStart first learns from easy states, and over time, starts learning from harder states until it masters a task to the most effective of its ability. It requires 100x fewer simulated demonstrations to learn how you can solve a task in simulation than what’s normally needed when learning from real world examples for a similar purpose.
The robot achieved a hit rate of over 98% on plenty of different tasks in simulation, including reorienting cubes with a certain color showing, tightening a nut and bolt, and tidying up tools. Within the real-world setup, it achieved a 97% success rate on cube reorientation and lifting, and 64% at a plug-socket insertion task that required high-finger coordination and precision.
