Embedded Files
Why should robots grasp like a humans?
Why should robots grasp like a humans?
Humans grasp objects with incredible ease. We can easily pick up one popcorn from a bowl or a single nut from a container, even without looking while watching a movie. This may sound spurring at first if we consider the chaotic motion of popcorn in a pile. Humans developed skills to simplify their lives. So we don't need to look or think about the object's motion actively while grasping from piles.
Humans grasp objects with incredible ease. We can easily pick up one popcorn from a bowl or a single nut from a container, even without looking while watching a movie. This may sound spurring at first if we consider the chaotic motion of popcorn in a pile. Humans developed skills to simplify their lives. So we don't need to look or think about the object's motion actively while grasping from piles.
To transfer such a skill to a robot, we need to understand what phenomena humans leverage during grasping. Hence, we need to study piles' behavior when grasping from it and extract a motion pattern that emerges consistently. We identified a motion pattern in piles, called Granular Environmental Constraint, and transferred the skill by designing simple grasp strategies. Our grasp strategies, just like humans, doesn't require to visually perceive individual objects in a pile and the robot executes simple straight motion to grasp an object robustly.
To transfer such a skill to a robot, we need to understand what phenomena humans leverage during grasping. Hence, we need to study piles' behavior when grasping from it and extract a motion pattern that emerges consistently. We identified a motion pattern in piles, called Granular Environmental Constraint, and transferred the skill by designing simple grasp strategies. Our grasp strategies, just like humans, doesn't require to visually perceive individual objects in a pile and the robot executes simple straight motion to grasp an object robustly.
What did we learn from our empirical study of human-like open-loop grasping?
First, we learned that studying a higher level motion in a pile (rather than an individual object's motion) can indicate useful Environmental Constraints (EC). In the case of piles, the granular EC effectively separates and supports objects for grasping.
Second, we don't need to mathematically model the interaction between the robot and objects to leverage the EC for grasping. Modeling the motions for a given pile would be difficult. This is because we can't see all objects, and we don't know the object's physical properties accurately. On the other hand, the robot can rely on physics without "knowing the math" because psychical laws govern the observed behavior, and thus, the behavior manifests consistently.
We analyzed our simple grasping strategies both in simulation and with a real robot and presented exciting findings in our publication presented at ICRA2021. Below, you can find a per-recorded presentation and the publication's reference:
Publication reference: Előd Páll and Oliver Brock, "Analysis of Open-Loop Grasping From Piles ", The 2021 International Conference on Robotics and Automation (ICRA 2021), 2021 (pdf link will be available shortly)
Google Sites
Report abuse