Artificial Intelligence & Machine Learning

Artificial Intelligence & Machine Learning: Awarded the Best Innovative Hardware for AI & ML Testing

The Ultimate Testing Ground for AI Technology

You can train our human-like Shadow Dexterous Hand to manipulate physical objects, and complete goal-orientated tasks intuitively and autonomously using machine learning and artificial intelligence.

Our Shadow Dexterous Hand has 20 degrees of freedom (the highest available on the market), allowing it to manipulate objects with unprecedented dexterity. Coupled with over 100 proprioception sensors to provide the data AI systems need to learn and achieve a range of dexterous tasks, at speed and scale.

Researchers have used neural networks to train our robot Hand in simulation using the trial-and-error principle (reinforcement learning). The data is then transferred to our robot Hand so that it can perform the desired action in a real-life setting and in real-time. Since our system has no prior information, it removes bias and enables researchers to explore the system freely for revolutionary results.

There are other ways to achieve nimble tasks with autonomy using our Hand and machine learning, such as learning by demonstration. We encourage AI researchers and experts to get in touch to discuss their needs.

Shadow is a proud winner of the AIconics Award for Best Innovation in AI Hardware 2019

 

Contact us for an AI demo, enquiries, or a purchase.

 

Case Studies:

OpenAI

We’ve worked with OpenAI, founded by business tycoons, Elon Musk and Sam Altman to advance research within AI and machine learning using the Shadow Dexterous Hand. OpenAI researchers take our hardware and explore machine learning by creating a robotic system called DACTYL in which a virtual robotic hand learns through trial and error. These human-like strategies are then transferred to the Shadow Dexterous Hand in the natural world, enabling it to grasp and manipulate objects efficiently, no fine-tuning needed!

OpenAI researchers also demonstrate a self-teaching algorithm that allows Shadow’s robot Hand to manipulate a cube with uncanny skill by practising for the equivalent of a hundred years inside a computer simulation – though only a few days in real-time! They describe it best on their website with some fantastic footage, also featured on BBC News and the front page of the New York Times.

 

People are able to perform a wide range of dexterous manipulation tasks in a diverse set of environments, making the human hand a grounded source of inspiration for research into robotic manipulation – OpenAI.

 

Human Brain Project (HBP)

We’re proud to have collaborated with HBP. This organisation aims to put in place a cutting-edge research infrastructure that will allow scientific and industrial researchers to advance our knowledge in the fields of neuroscience, computing, and brain-related medicine. HBP successfully integrate and simulate the Shadow Hand on the HBP Neurorobotics Platform which connects a physics simulator to a variety of neural networks (or brains).

Our hardware allows for quick experimentation, testing and reconfiguration of neural networks, removing the need for time-consuming physical setups. Each Shadow Hand joint can be individually linked to both the input and/or output of the brain, giving the user a considerable level of control over the link between Hand and neural network. Contact based feedback is also forwarded to the brain.

You can read more about our contribution to bridging the gap between neuroscience and robotic communities on HBP’s website.

 

Google Brain

Google Brain pursue a novel robot task planning technique involving deep dynamics models or DDM. Researchers can manipulate multiple objects using the Shadow Hand with just four hours of real-world data.  One of the most difficult real-world manipulation challenges involved rotating two Baoding balls (Chinese relaxation balls) around the palm without dropping them. Remarkably, the researchers’ models can solve it, using just 100,000 data points’ (or 2.7 hours) worth of data.

In a separate experiment, the team repurpose models trained on the Baoding task to accomplish other tasks without additional training, including moving a single ball to a goal location in our robotic Hand and performing clockwise rotations instead of learned counter-clockwise ones. You can read all about the impressive results in this article.

 

Pharmaceutical Settings

We have a keen focus on the pharma industry, and our robot Hand can learn from demonstration, which can be revolutionary for this sector. It begins with our teleoperation system where an operator controls our Hand via a haptic glove to carry out mission-critical tasks. The hardware can log the trajectory, grasp forces and various other data which can then be analysed by AI experts who can programme the Hand to repeat the behaviour autonomously.

 

If you’re interested in using our Shadow Hand as an AI and machine learning testing tool, feel free to reach out.

Put me in touch