What Is the Future of Robotic Machine Learning Research?
- Apr 22, 2024
- 5 min read
The Short Version
Focus your research on the arms-and-hands modal subset, the most sought-after and versatile human toolset.
Choose reliable, cost-effective hardware like the Aloha Research Kits to iterate faster than the $400,000 PR2 era.
Match human arm motion with 6 Degrees of Freedom manipulators such as the ViperX and WidowX arms.
Teleoperate follower arms via the leader-follower design to naturally record joint positions, speed, and timing.
Capture time-encoded video from multiple cameras alongside joint data as primary machine-learning inputs.
Scale training across multiple low-cost kits like Aloha Stationary and Aloha Mobile to build a large data pool.
Pass the data pool of success-and-failure experience between models to accelerate development and iteration.
Who this is for
Robotic machine learning researchers
Robotics R&D engineers
University and lab research teams
Teleoperation and manipulation developers
Commercial and home robotics innovators
The future of robotic machine learning research is being driven by data-driven intelligence that can advance robotics from basic mimicry to intuitive actions on command. Robotic Machine Learning has been gaining a lot of attention lately. Various advancements, such as Nvidia's Project GR00T and Stanford's Aloha Project, have left the world wondering what the future holds.
While the bipedal androids Tesla, Boston Dynamics, and others have been creating are shiny and awe-inspiring, the research behind how they are engineered and programmed is a lot less flashy. It's worth highlighting how researchers and engineers are tackling the challenge of developing and advancing robotic machine-learning models for the future.
Two kinds of robots: what's the difference?
Let's categorize robotics and machines into two groups: those that imitate human mobility and those that perform tasks beyond human capability.
Type | Examples | Design |
Beyond human capability | Cranes, dump trucks, space satellites | Built for specific applications; not general-purpose, with a limited range of intended functions |
Imitate human mobility | Android-like robots and manipulators | More versatile and complex to design for general-purpose applications |
Machines that mimic human movements tend to be more versatile because humans are capable of mastering a wide range of skills: building houses, baking bread, swimming, running, sewing, driving, gesturing, exercising, and more.
Why did affordable hardware change robotics research?
In the early years of robotic machine learning research, hardware was a major expense — a barrier that reliable, cost-effective hardware has since removed.
One of the first android-like robots created for research was the PR2 by Willow Garage. It cost an estimated $400,000, and only 11 labs in the world could afford it.
Fast forward to today, and reliable, cost-effective hardware has accelerated the pace researchers can develop and iterate. The same can be said for the availability of specific modal-optimized research kits like the Aloha Research Kits from Trossen Robotics. The hardware can be designed for a specific subset of human modes of movement, such as arms and hands.
Why do researchers focus on arms and hands?
Our arms and hands are responsible for a significant portion of the daily tasks we undertake. They are our most versatile tools, arguably second only to speech and communication. Because we use them more than almost any other appendage, they are one of the most sought-after modal subsets researchers focus on.
Human arms require 6 Degrees of Freedom to position the wrist and orient the palm — the same number of Degrees of Freedom in the ViperX and WidowX Robotic Manipulator Arms used in the Aloha Research Kits.
How do the Aloha Research Kits train models?
The Aloha Research Kits implement a leader-follower arm design. A researcher holds and manipulates the leader arms to teleoperate the follower arms with grippers to perform training tasks. As mentioned, the arms are designed to mimic the joints of a human arm, so the researcher can move and position the arm and gripper naturally.
The system then records the joint positions and speed within the joint space, continuously logging every step in the training episode for the user-defined time. Combined with time-encoded video feeds from multiple cameras, you have the primary inputs for a reliable machine-learning model capable of recreating tasks with a high degree of success. The Trossen SDK gives researchers the software layer to capture and replay this data across their hardware.
Why does general-purpose robotics need so much training data?
Researchers need to train models to perform as many tasks as a human can — multiple times, and in multiple ways — to achieve general-purpose functionality. Think of the number of ways you could peel a banana, depending on where it is positioned relative to you, its size, shape, ripeness, and more.
Tasks also require training to be distributed over a number of hardware instances to achieve the sheer volume of data needed. A single machine or individual operator cannot be relied upon for thousands of hours of training. Natural language models like ChatGPT reportedly took approximately 1 million hours of training to achieve general-purpose functionality from thousands of participants.
How do low-cost kits build a training data pool?
Researchers can build an extensive training data pool using multiple low-cost research kits like the Aloha Stationary and Aloha Mobile. The larger the data pool, the more “experience” the model has to draw from — both success and failure. Like humans, this data pool of experience can be passed on from one model to another, allowing rapid development and iteration.
What's next for robotic machine learning research?
The development of general-purpose robotic machine-learning models is rapidly evolving. In the coming months and years, it will move out of the lab and into the consumer and commercial markets, where it will be used for applications in the home, the factory, and many other areas.
We at Trossen Robotics are excited to be part of the R&D community by providing the hardware researchers and engineers need to tackle these challenges.
---
“PR2”. Robots Guide. https://robotsguide.com/robots/pr2
H. Kim, L. M. Miller, N. Byl, G. M. Abrams and J. Rosen, "Redundancy Resolution of the Human Arm and an Upper Limb Exoskeleton," in IEEE Transactions on Biomedical Engineering, vol. 59, no. 6, pp. 1770-1779, June 2012, doi: 10.1109/TBME.2012.2194489. https://ieeexplore.ieee.org/document/6182581
Jonathan Vanian and Kif Leswing, March 13, 2023. “ChatGPT and generative AI are booming, but the costs can be extraordinary”. CNBC. https://www.cnbc.com/2023/03/13/chatgpt-and-generative-ai-are-booming-but-at-a-very-expensive-price.html
Frequently Asked Questions
What is the future of robotic machine learning research?
It is data-driven intelligence advancing robotics from basic mimicry to intuitive actions on command, moving out of the lab and into consumer and commercial markets like the home and the factory in the coming months and years.
Why do researchers focus on arms and hands?
Our arms and hands are our most versatile tools, arguably second only to speech and communication, and are responsible for a significant portion of daily tasks, making them one of the most sought-after modal subsets.
How much did early research hardware cost?
The PR2 by Willow Garage, one of the first android-like research robots, cost an estimated $400,000.00, and only 11 labs in the world could afford it.
How many Degrees of Freedom do the Aloha Research Kit arms have?
Human arms require 6 Degrees of Freedom to position the wrist and orient the palm—the same number found in the ViperX and WidowX Robotic Manipulator Arms used in the Aloha Research Kits.
How do the Aloha Research Kits record training data?
They use a leader-follower arm design where a researcher teleoperates the follower arms, and the system records joint positions and speed in joint space, logging each step with time-encoded video from multiple cameras.
How much data is needed for general-purpose functionality?
Training must be distributed across many hardware instances for sheer volume; for comparison, natural language models like ChatGPT reportedly took approximately 1 million hours of training from thousands of participants.
How does Trossen Robotics support this research?
Trossen Robotics provides the reliable, cost-effective hardware researchers and engineers need, including modal-optimized kits like the Aloha Stationary and Aloha Mobile, to tackle these challenges.
Sources
_Citations preserved from the original article._
Comments