Embedded Files
A Real-time Solution to Slosh-Free Robot Trajectory Optimization
Introduction
"Jenkins! You have too much coffee in your mug!”
The general expression to say there is nothing worse than spilling coffee/hot beverages on oneself or others is the starting point of our no-spilling robotics project. The work is originally focused on service robotics, particularly in the context of Geriatronics and elderly care scenarios whereby it is critical for a robot to serve food and liquids, such as hot soups and/or beverages, to an elderly while being reactive and guaranteeing not spilling. Furthermore, there are also many applications in manufacturing and industry where transporting liquids is nowadays performed in a slow and conservative manner, which degrades performance and increases costs.
Robots also don't like spilling coffee
Intuitively, one can reason about the sloshing problem as being caused by the reaction forces exerted on the liquid from the inner walls of the container due to higher-derivatives impulse-like motions, for instance, high jerks and accelerations. Therefore, a slosh-free motion can be found when there are no lateral reaction forces nor external torques acting on the container. This case arises when the total reaction force is aligned with object’s vertical direction and its line of action passes through the object’s center of mass. We take inspiration from three centuries old Moroccan tray and a market solution for coffee/liquid transportation - a device known as the SpillNot - both designed keeping the same idea in mind.
Continue reading to learn more about our framework, or checkout these additional resources :
Read the technical details of the modeling and control architecture in our first paper, the Spillnot Trajectory Optimization [IROS 2022]
Read the technical details of shared-autonomy control for Slosh-free Teleoperation [Under Review]
View the motion generation implementation in our Git page (To be Updated)
Demo Video
Demo Video
Robot Reaction in less than a millisecond
Robot Reaction in less than a millisecond
In the SpillNot project we are interested in real-world applications. For example, if a person is served a glass of tea or a bowl of soup by a robot in a nursing home or service scenario, the robot must be able to move and act promptly. And, most importantly, if the person reacts abruptly, the robot also needs aggressive reactive maneuvers to escape any collision, and this needs to be done in real-time. The same goes for any real-world industrial scenario. Our online control architecture is capable of addressing any safe fluid transportation (in a 1 KHz frequency).
The slosh-free trajectory, further detailed in our SpillNot Trajectory Optimization, stems from a linear ODE approximation of the nonlinear equations of motion of a spherical pendulum. It relies on a point-mass approximation of liquid bodies, which is also investigated in the original paper. The limits for the approximation and the error are also detailed therein.
Further implementation details can be found in the links above.
Teleoperating SpillNot
Teleoperating SpillNot
The SpillNot robot motion generation strategy is built for real-time usage and enables users to transport liquids in ways not possible before — even with non-prehensile manipulation features (e.g. transporting a tray with liquid with slosh-free guarantees). Particularly, we are interested in leveraging human cognitive abilities with spillnot’s capability to exceed human-level performance for the task. Take, for instance, the task of transporting hazardous liquids, retrieving liquid samples in industrial facilities or disaster areas, or even just manipulating a cup with hot beverages filled up to the surface tension. Whilst the manipulation task may be hard – and even dangerous – for humans, our cognition, and reaction skills are critical to defining and adapting the task on the fly.
In this sense, spillnot motion generation may provide the human with complementary skills for the task, which can be completed through teleoperation.
The pipeline for SpillNot teleoperation strategy
The proposed control architecture has three main components: (i) a time-to-bound safety mapping block, which feedbacks virtual forces to the leader and modifies the input commands to satisfy a set of constraints—for instance, joint-limits, singularities, and task-space limits; (ii) a constrained LQG, which acts as the motion generator and ensures the resulting trajectory—in terms of pose, twists, and accelerations—satisfies the conditions for the slosh-free equations of motion; and (iii) a task-space motion controller, which yields the control signal and updates the states' information to the Kalman filter and to block (i).
Releasing SpillNot robot motion generation to the world
Releasing SpillNot robot motion generation to the world
IROS 2022- with Franka Emika - Exhibition
IROS 2022- with Franka Emika - Exhibition
We invite all the roboticists that are attending the IROS conference to visit Franka Emika's booth at the exhibition hall to see and play with our interactive demo. You can teleoperate with a robot arm to achieve above-human no-spilling transportation !
SpillNot teleoperation exceeding human performance
IROS '22 demo setup at Franka Emika
Automatica Munich_i Exhibitor
Automatica Munich_i Exhibitor
The SpillNot robot motion generation has also been presented at the Automatica Munich_i, the world largest fair about robotics and automation that happened in June 21-24, 2022. At the Automatica, our framework was exposed to the general community, experts, academics, and the industrial community for 4 consecutive days without ever spilling !
Contact
Contact
Page updated
Google Sites
Report abuse