Oregon State’s Use of Bird Models in Robotics

Chris Moehle leads Pittsburgh’s The Robotics Hub and guides a fund dedicated to enabling next-generation robotics advancements. Among the portfolio companies funded through Chris Moehle’s organization is Agility Robotics, which is in the process of creating legged robots that have perception sensors and a pair of arms built in.

Before it was a company, Agility was a creation of the Dynamic Robotics Laboratory at Oregon State University (OSU). Led by Jonathan Hurst, who is now Agility Robotics’ Chief Technology Officer, the Dynamic Robotics Lab worked with a number of national and international collaborators in understanding the mechanical basis of effective movement. This collaborator list includes the Royal Veterinary College of the University of London. Together, OSU and the Royal College observed the mechanics of birds walking and running within laboratory and field settings. A central focus is on how these “feathered machines” are able to efficiently and nimbly travel between two points with relatively little “central processing”.

One experiment had a guinea fowl run down a track that contained a pothole half a leg deep (which it was unaware of), covered by tissue paper. Without the bird’s brain having time to sense or react, the leg itself adjusted to the drop at the moment it encountered it. A key insight was that designing a robotic leg and later programming it to move effectively is a backwards way of going about the design, as effective dynamics should – and almost must – be built into the robot’s core mechanical structure.

This is important to consider both when the robot is moving, and when it is not. By being mindful of the passive and active dynamics of the machine, software controls become simpler and more approachable. As a result, you get robots that are more robust and controllable in uncertain real-world circumstances. As a nod to this insight from our avian friends, and the resulting core design principles for robots, Agility named their first robot “Cassie” – an affectionate abbreviation of New Guinea’s flightless cassowary.

Agility Robotics’ Next-Generation Legged Robots with Arms and Sensors

Agility Robotics
Image: AgilityRobotics.com

Guiding The Robotics Hub in Pittsburgh as Managing Director, Chris Moehle leads a pioneering fund that seeks out growing companies focused on the area of emerging robotics. Among the companies funded by Chris Moehle’s firm is Agility Robotics, which builds on existing technologies that allow advanced humanoid robots to walk, climb, slog, and jump.

A major issue is that legged robots—even those developed by leading companies such as Boston Dynamics—are still unable to match the robustness and efficiency of animal and human movement. With this in mind, researchers at Dynamic Robotics Laboratory and Oregon State University focused on better understanding legged locomotion principles and applying the findings to robotics. As a result, the bipedal robot Cassie was launched in 2017 and sold to a number of research groups.

The forthcoming Digit robot builds on Cassie with a similar leg structure, but with the addition of arms and perception sensors that will initially be used to increase stability, and ultimately to enhance manipulation capacities. The result aims to be the first robot capable of thriving in real-world environments built for humans. Working collaboratively with humans, Digit promises to help reduce injury, damage, and accidents. Ultimately, this creates a better environment and economy for man and machine alike.

A Hybrid of Machine Learning and Quantum Computing Emerges

Serving as managing director of The Robotics Hub, Chris Moehle guides a Pittsburgh-based company focused on funding advancements in the area of robotics. One area in which The Robotics Hub’s portfolio companies have a strong interest in (as a way of creating robots with superior decision-making capacities) is the intersection between machine learning and quantum computing.

Areas of emphasis within this hybrid field include the use of nascent quantum computers to expedite machine learning algorithms, with the ultimate goal being to employ the smallest possible computing system in interpreting and understanding large-scale datasets.

This area of research has its impetus in HHL, a quantum algorithm developed in 2008 that has the ability to find answers to vast linear algebra problems with many degrees of freedom at a potentially faster speed than traditional supercomputers. A key advantage of HHL over standard machine learning algorithms is that such a system can generate purely random numbers. One limitation is that quantum machine learning algorithms have thus far been designed as “frameworks for algorithms” instead of posing a classical problem that needs to be solved with a logically derived answer.

In sum, this provides a new way to solve problems. In areas where there is no classical solution – like areas of materials science and medicine – research, development, and implementation can be somewhat straightforward. However, in areas where classical computing is currently performing – such as navigation and optimization – research, development, and implementation must be somewhat more nuanced. An ideal solution allows quantum to be “layered” with traditional approaches allowing the stability of classical algorithms to coexist with the enhanced functions quantum promises.