Lessons from the Amazon Picking Challenge

Lessons from the Amazon Picking Challenge

Jul 6, 2016 @ 18:41 | 

Amazon Picking Challenge is to build their own robot hardware and software that can attempt simplified versions of the general task of picking items from shelves. The challenge combines object recognition, pose recognition, grasp planning, compliant manipulation, motion planning, task planning, task execution, error detection and recovery. To get the job done, the robot needs to deal with unsolved problems for automation. The Amazon Picking Challenge tested the ability of robotic systems to fulfill a fictitious order by autonomously picking the ordered items from a warehouse shelf.

Team RBO, TU Berlin, Germany won the first Amazon Picking Challenge at ICRA 2015. 25 teams from Europe, USA and Asia competed for it. With a total of 148 points team RBO (Robotics and Biology Laboratory) was able to secure the first place.

On the basis of experience gained in journey of building Picking robotic system and becoming winner of first Amazon Picking Challenge RBO team published “Lessons from the Amazon Picking Challenge: Four Aspects of Building Robotic Systems” in RSS 2016. Researchers present three main conclusions from own system-building experience:

  1. Robotic systems can be characterized along four key aspects. Each of these aspects can be instantiated by selecting from a spectrum of approaches.
  2. To develop a shared understanding of system building, i.e. which region of each spectrum most adequately addresses a particular robotic problem; we should explore these spectra and characterize systems based on them.
  3. For manipulation in unstructured environments, researchers believe that certain regions of each spectrum match the problem characteristics most adequately and should therefore be examined by roboticists with increased emphasis.

And to build a robust and reliable robotic system for picking challenge, we have to take care of these four aspects:

  1. Modularity vs Integration: In robotics, the behavior of the entire system determines success, not the performance of individual modules. Still, a high degree of modularity allows breaking down problems into simpler sub-problems, which is especially useful when the overall problem is too complex to solve. Wrong modularization, however, can make solving problems unnecessarily difficult. Until you fully understand which modularization is most adequate for manipulation in unstructured environments, try to build tightly integrated systems and constantly revise your modularization.
  2. Computation vs Embodiment: Robot behavior results from the interplay of computation (software) and embodiment (hardware). Computation is a powerful and versatile tool but adapting the embodiment sometimes leads to simple and robust solutions. Researchers suggest that in manipulation, one should consider alternative embodiments as part of the solution process, so as to most synergistically match software and hardware.
  3. Planning vs Feedback: Planning performs search in a world model, leading to verifiable solutions. Feedback from physical interactions, on the other hand, reduces uncertainty and allows to find local solutions without expensive computation. Researchers thus suggest to use planning only when necessary and explore the use of feedback as an alternative when the manipulation task does not require global search.
  4. Generality vs Assumptions: For robotics research, finding general solutions is highly desirable because they apply to a wide range of robotic tasks. However, solving the most general problem might be unnecessary or even unfeasible. Researchers suggest to search for reasonable and useful assumptions that aid solving manipulation problems in unstructured environments.

Researchers also suggest that these aspects are not novel and certainly will not surprise the robotic practitioner. However, what should come as a surprise is the sparsity with which the corresponding spectra have been explored by our community and how rarely these aspects are used explicitly to characterize robotic systems.


  • Reference: “Lessons from the Amazon Picking Challenge: Four Aspects of Building Robotic Systems”, Clemens Eppner Sebastian H¨ofer Rico Jonschkowski Roberto Mart,  RSS 2016 [paper]
  • Image: Robotics and Biology Laboratory, TU Barlin, Germany

One Comment

  1. I see your page needs some fresh & unique articles. Writing manually is
    time consuming, there is tool for this task.
    Just search in gogle for – Fejlando’s tips

Leave a Reply

Your email address will not be published. Required fields are marked *