ABB and Psyonic Put Prosthetic-Hand Data to Work in Cobots
Robotics firm ABB is collaborating with bionic-hand manufacturer Psyonic on whether data from human manipulation of a robotic hand can translate into better object handling by collaborative industrial robots. It’s a novel technical approach with significant potential implications for the advancement of robotic manipulation: Can manipulation data generated through human use of a prosthetic hand be used to improve robot manipulation?
The joint research will focus on using a robotic hand as an end-effector for ABB’s GoFa collaborative robot. It’s worth noting that Psyonic’s Ability Hand can function both as a human end effector and as a robotic end-effector. Therefore, researchers could potentially collect touch, grip, and motion data when the robotic prosthetic is worn by humans, then apply that data to robotic handling scenarios.
This particular approach is interesting from an engineering standpoint, as dexterous manipulation continues to be among automation’s most stubborn challenges. Conventional robotic grippers can excel in situations where there are consistent parts and consistent presentation, but become inefficient in scenarios where there’s fragility, complexity, or variation. Those conditions are precisely the type of handling situation one encounters in mixed manufacturing processes, logistics flows, packaging lines, laboratories, and even some aerospace-adjacent assembly and kitting applications.
ABBAutomation has long recognized the opportunity presented by developing advanced robotic handling capabilities for manipulating delicate, irregular, and variable objects, said Dr. Marc Segura, President of ABB Robotics Americas, in a press release. The company says it will use GoFa as the test platform, presumably due to its ability to perform very precise movements and record small changes in grip force and finger positioning.
In other words, this isn’t an experiment where the robot is merely holding the prosthetic. Instead, the test setup is designed to see if human data can produce repeatable robotic manipulation of complex objects. To make that possible, the research is centered on the sensorization of the Ability Hand’s articulated multi-fingered hand architecture, which was created by Psyonic. As a prosthetic hand, it can detect touch, pressure, and multi-joint finger movement to allow natural interaction with daily objects. From the perspective of a robotic application, that sensorized and articulating structure represents an interesting avenue for improved dexterity.
As a bionic prosthesis, explains Dr. Aadeel Akhtar, Founder and CEO of Psyonic, the Ability Hand uses machine learning algorithms to understand movement, contact, and grip force to assist the user’s intuitive motor planning in handling everyday objects. This allows us to capture “high-fidelity real-world data on movement, contact and grip force” to train robotic systems. That’s the key technical takeaway here. As Akhtar puts it, “Dexterous manipulation is ultimately a data challenge as much as a hardware challenge.”
From that angle, the challenge can be framed as follows: How can manipulation be taught to robots via data? To explain, let us consider how ABB frames its robotics business. As president Marc Segura puts it, human dexterity and the inherent capability to intuitively manipulate a wide variety of objects remains a hard problem. Consequently, end-effectors, tactile sensors, control loops, and data acquisition must be thought of holistically as a system.
That particular perspective represents an ongoing evolution of robotics designs. Manufacturers have focused heavily on improving arms and controllers over the years, but there are still many difficult-to-implement tasks in a cell. That’s particularly true for mixed manufacturing environments with highly variable component sizes. Many robotic solutions localize objects perfectly, yet struggle to grasp fragile and compliant objects. Thus, improved dexterity isn’t simply a question of putting additional fingers on the end effector. Rather, it hinges on data.
In other words, better manipulation requires not just an articulating hand design, but the ability to sense, apply force, and make fine adjustments in positioning based on real-time inputs. That requirement in turn presents another challenge. As shown by ABB’s controller architecture – which includes its OmniCore product family – real implementation is contingent upon not only the articulating end effector. Issues such as safety functionality, IO ports, field bus connectivity, vision system integration, and tool control also come into play, particularly when working in a limited manufacturing cell.
AABB’s definition of Autonomous Versatile Robotics further illustrates the point. Robots that have the capability to sense, reason, move and manipulate objects in dynamic environments are the foundation of “Autonomous Versatile Robotics,” according to the company. Note that the concept is defined with regard to capabilities, not specific implementations.
Thus, ABB views the current challenge of robotic manipulation as one of expanding dexterity while ensuring continued reliability of handling. The solution to this challenge involves the combination of tactile sensing, control, and machine learning of data acquired through manipulation.
In other words, we’re witnessing an evolutionary step in robotics design. In recent years, robotic arms and controllers have been optimized significantly, but many difficult tasks remain in the field of grasping and manipulation. Improving human dexterity by combining all these elements into a single robot-hand setup is therefore one of the biggest opportunities right now.
The initial benefit of this collaboration is clear: Using data from human manipulation, collaborative robots might be able to apply appropriate pressure and correct finger positioning for delicate or otherwise difficult-to-handle objects. For high-mix manufacturers in the U.S., this represents an important breakthrough in industrial robotics.
By Jonathan Barrett — Editor for AMI’s future mobility and autonomous systems section, with two decades covering robotics, e-mobility, drone-vehicle convergence, and transport mechanical systems.
