Last week, the SecondHands project unveiled the first prototype of its collaborative robot, which is intended to work in online grocer Ocado Technology‘s warehouses. The project is a recipient of Horizon 2020 funding from the EU and involves partners at multiple research institutions across Europe in addition to the Ocado team.
U.K.-based Ocado already has highly automated facilities, but SecondHands’ ARMAR-6 is noteworthy because of its humanoid design and the intention for it to work alongside people. Robotics Business Review asked the Ocado team for more details about the project.
In the two years since we first wrote about SecondHands, what were the biggest challenges as the project developed?
There are many challenges on the project. All component parts from the different partners are state-of-the-art research, and so each has presented its own challenges. Bringing these separate components together to work as a single coherent whole is a challenge in itself, especially since much of this work had to be performed in parallel with the robot being constructed.
How much funding did the EU provide through the Horizon 2020 program?
The total cost of the project has been 6.9 million euros [$8.44 million U.S.]. The EU contribution is 5.9 million euros [$7.2 million], and the remaining has been provided by the Swiss government directly to the Swiss partner EPFL, due their EU Horizon 2020 membership at the start of the project. More information about the funding can be found here.
While each of the research institutions operated independently, how did the Ocado team coordinate the elements to make sure they’d be interoperable?
Ocado’s role in the project is the project coordinator and leading the work package for integration. However, every partner also contributes on the integration to adapt and configure their developed components to operate as a cohesive robot system. While some of the research is independent, other parts have been integrated early in the project, such as the vision system for object pose estimation and the grasping pipeline.
Other research output has been adapted for integration during the past year of the project in preparation for the completed ARMAR-6 robot. An integration framework, based on containers, allowed components to remain independent and to be individually tested, while combining the power of vision, knowledge and planning, natural language processing, human-robot bi-manipulation, and robot control. The integrated robot system was being tested before the robot was completed.
Your video (above) states that you expect to conduct a demonstration in about three weeks — why such a short timeline? Do Ocado and Horizon 2020 still expect production models by 2020?
We always knew the design and build of a complete humanoid robot was very ambitious in the timescale. The strategy for integration meant we could test the interfaces between the research components and perform basic functional tests in simulation.
The real world is much more complex than we could model in a simulator, such as noise or uncertainty from sensors, real-time control of all the motors and actuators on the robot. The short timeline has arisen due to our commitment to complete our objective of the first demonstration of the prototype ARMAR-6 performing a straightforward maintenance assistance task coincident with the second review period for the EU.
We are very happy to achieve the rapid integration and testing of all the components thanks to the hard work of all the project team members.
What sorts of human-like tasks does the Ocado team want SecondHands to tackle first?
The tasks that we want the SecondHands robot to get involved in are relatively simple maintenance tasks, like handing over tools or receiving component parts during the disassembly of pieces of automation. The idea is to make maintenance technicians more efficient by providing them with a a second pair of hands — hence the project title — not to usurp their function. We will still be relying on the skill and experience of the maintenance technician.
It is probably worth pointing out that the ambition of the project is not so much about the tasks that the robot gets involved with, but more the ability of the robot to understand where in a maintenance task the technician is and know what assistance from its behavioral repertoire would be appreciated at that point. We want a robot system that displays cognitive abilities.
Does Ocado face similar worker shortages in the U.K. to those experienced by certain industries in the U.S. and Asia?
There is an estimated shortfall of 20,000 engineering graduates a year in the U.K., and there have been growing reports from industry that the skills shortage is having a significant impact on productivity.
While automation can definitely improve productivity, we are also investing in programs such as Code for Life to ensure that enough young people in the U.K. and worldwide become interested in an engineering career.
Who will be manufacturing SecondHands? Is there a plan for commercialization beyond use by Ocado?
The SecondHands project is a project conducting fundamental research. Even by the end of the project, the SecondHands robotic system will still be a prototype device. We are not expecting it to be immediately deployable for commercial purposes.
It is the nature of fundamental research projects that exactly what we will have achieved by the end of the project is uncertain. We will evaluate what we consider the best next steps for the technology developed as we approach the end of the project and have a better appreciation of its scope.
Has the Ocado team been following the progress of other humanoid robot research such as Toyota’s T-HR3? What are some similarities and differences?
We are aware of other research and commercial projects, but we believe the AI capabilities of the SecondHands robot are unique.
Furthermore, while some industrial robots tend to position themselves in ways that are uncomfortable to humans, we really wanted to design a robot that can work seamlessly with other humans rather than get in the way. Research shows people are more comfortable working with both the morphology and behavior that they expect to see.