Michael Bombile Bosongo

Related publications (5)

Bimanual grabbing and tossing of packages onto trays or conveyor belts remains a human activity in the industry. For robots, such a dynamic task requires coordination between two arms and fast adaptation abilities when the tossing target is moving and subject to perturbations. Thus, this paper proposes a control framework that enables a bimanual robotic system to grab and toss objects onto a moving target. We develop a mixed learning-optimization method that computes the tossing parameters necessary to achieve accurate tossing tasks. Hence, we learn an inverse throwing map (a closed-form solution of the inverse non-linear throwing problem) that provides minimum release velocities of the object for given relative release positions. This map is embedded into a kinematics-based bi-level optimization that determines the associated feasible release states (positions and velocities) of the dual-arm robot. Additionally, we propose a closed-form modeling approach of the robot’s tossable workspace (set of all positions reachable by an object if tossed by the robot) and use the model to predict intercept or landing locations that yield high probabilities of task success. Furthermore, we employ dynamical systems to generate the coordinated motion of the dual-arm system and design an adaptation strategy to ensure robustness of the interception in the face of target’s perturbations in speed or location. Finally, we validate experimentally the framework on two 7-DoF robotic arms. We demonstrate the accuracy and robustness of the proposed approach. We also show its speed and energy advantages when compared to the traditional pick-and-place strategy.

2023

Bi-manual picking up of objects to toss them on a conveyor belt are dynamic manipulation activities generated daily in the industry. Such repetitive and physically demanding tasks are still done largely by humans for lack of similarly fast, precise, and robust bi-manual robotic systems. With nowadays booming of e-commerce, the needs for faster package handling solutions continue to increase. Hence, automation and robotics is the only viable solution as the current workforce cannot keep up with the growing industry demands. In current applications, however, robots usually use quasi-static approaches (with near zero relative velocities) to grab and release objects, mainly to avoid impacts. Thus, this thesis aims to develop dynamic alternative solutions to quasi-static manipulation approaches with the goal to accelerate object handling operations and improve their energy efficiency. It focuses on bi-manual (dual-arm) robotic manipulation for its potential to mimic humans' dexterity. To achieve its goal, the thesis considers different scenarios with two types of bi-manual systems: a biped humanoid robot and a pair of fixed base robot manipulators. The first part of the thesis starts by endowing the humanoid robot with balance and locomotion abilities needed for dual-arm cooperative tasks. It proposes a reactive locomotion controller that exploits the capture point dynamics to generate on-the-fly adjustable omnidirectional walking patterns that are consistent with balance constraints. In the second part, the whole body of the humanoid robot is controlled to accomplish bi-manual motion coordination and cooperative compliant manipulation tasks. The proposed approach relies on dynamical systems and exploits a shrinkable virtual object concept with its associated constraints to achieve robust coordination of the robotic hands with smooth transitions between non-contact and contact motion phases. The framework uses quadratic programming (QP) to generate constraint-consistent interaction wrenches that stabilize the grasp and perform the desired manipulation tasks. The third part of the thesis considers more dynamic interactions of the dual-arm system with the objects by allowing the grabbing and releasing of objects with non-zero relative velocities. Such an approach, besides speeding up the task, offers the possibility to expand the robot's reach beyond its physical boundaries. Thus, a unified coordination framework based on modulated dynamical systems is proposed for reaching, grabbing with impact, and tossing objects in one swipe. It is based on modulated dynamical systems that ensure motion continuity and robustness throughout the task. The last part of the thesis extends the dynamic framework by enabling the precise tossing of objects onto a moving target carried by a conveyor belt. It uses a learned inverse throwing map within a kinematic-based bi-level optimization to determine feasible tossing parameters required by the task. Moreover, it proposes and uses a model of the tossable workspace to determine intercept locations that yield a high probability of success. Finally, the proposed approaches are validated both in simulation and on real robotic platforms: a humanoid robot iCub and a pair of KUKA LBR (IIWA7 and IIWA14) robots. Kinetic comparisons with the classical pick-and-place strategy are conducted and the results show that the proposed swift pick-and-toss reduces the task duration and the energy expenditure.

EPFL2022

Picking up objects to toss them on a conveyor belt are activities generated on a daily basis in the industry. These tasks are still done largely by humans. This paper proposes a unified motion generator for a bimanual robotic systems that enables two 7 degrees of freedom robotic arms to grab and toss an object in one swipe. Unlike classical approaches that grab the object with quasi-zero contact velocity, the proposed approach is able to grasp the object while in motion. We control the contact forces prior to and following impact so as to stabilize the robots’ grip on the object. We show that such swift grasping speeds up the pick-and-place process and reduces energy expenditure for tossing. Continuous control of reach, grab and toss motion is achieved by combining a sequence of time-invariant dynamical system in a single control framework. We introduce a state-dependent modulation function to control the generated velocity in different directions. The framework is validated in simulation and on a real dual-arm system. We show that we can precisely toss objects within a workspace of 0.2 by 0.4 square meter. Moreover, we show that the algorithm can adapt on the fly to changes in object location.

2022

Learning to Hit: A statistical Dynamical System based approach

Aude Billard, Michael Bombile Bosongo, Harshit Khurana

This paper proposes a manipulation scheme based on learning the motion of objects after being hit by a robotic end-effector. This allows for the object to be positioned at a desired location outside the physical workspace of the robot. An estimate of the object dynamics under friction and collisions is learnt and used to predict the desired hitting parameters (speed and direction), given the initial and desired location of the object. Based on the obtained hitting parameters, the desired pre-impact velocity of the end-effector is generated using a stable dynamical system. The performance of the proposed DS is validated in simulation and and is used to learn a model for hitting using real robot. The approach is tested on real robot with a KUKA LBR IIWA robot.

IEEE2021

Capture-Point Based Balance and Reactive Omnidirectional Walking Controller

Aude Billard, Michael Bombile Bosongo

This paper proposes a capture-point based reactive omnidirectional controller for bipedal locomotion. The proposed scheme, formulated within Model Predictive Control (MPC) framework, exploits concurrently the Center of Mass (CoM) and Capture Point (CP) dynamics. It allows the on-line generation of the CoM reference trajectory and the automatic generation of footstep positions and orientations in response to a given velocity to be tracked, or a disturbance to be rejected by the robot while accounting explicitly for different walking constraints. For instance, in order to cope with disturbance such as a push, the proposed controller not only adjusts the position of the Center of Pressure (CoP) within the support foot, but can also induce at least one step with appropriate length allowing thus to maintain the stability of the robot. Finally, the proposed algorithm is validated through simulations and actual experiments on the humanoid robot iCub.

2017

EPFL2022