Vincent Padois

On-line regression algorithms for learning mechanical models of robots: A survey

With the emergence of more challenging contexts for robotics, the mechanical design of robots is becoming more and more complex. Moreover, their missions often involve unforeseen physical interactions with the environment. To deal with these difficulties, endowing the controllers of the robots with the capability to learn a model of their kinematics and dynamics under changing circumstances is becoming mandatory. This emergent necessity has given rise to a significant amount of research in the Machine Learning community, generating algorithms that address more and more sophisticated on-line modeling questions. In this paper, we provide a survey of the corresponding literature with a focus on the methods rather than on the results. In particular, we provide a unified view of all recent algorithms that outlines their distinctive features and provides a framework for their combination. Finally, we give a prospective account of the evolution of the domain towards more challenging questions.

Synthesis of complex humanoid whole-body behavior: A focus on sequencing and tasks transitions

We present a novel approach to deal with transitions while performing a sequence of dynamic tasks with a humanoid robot. The simultaneous achievement of several tasks cannot be ensured, so we use a strategy based on weights to represent their relative importance. The robot interacts with a changing environment, and the input torques are different depending on whether the robot performs tasks in a constrained state (e.g. in contact) or not. We develop a solution with smooth weights variations and transitional tasks which avoids sharp torque evolutions. In order to validate this approach, simulations are carried out on a virtual iCub robot which is assigned the realization of a complex mission involving various changing tasks.

Kinematic and dynamic model-based control of wheeled mobile manipulators: A unified framework for reactive approaches

The work presented in this paper aims at providing a unified modelling framework for the reactive control of wheeled mobile manipulators (WMM). Where most work in the literature often provides models, sometimes simplified, of a given type of WMM, an extensive description of obtaining explicit kinematic and dynamic models of those systems is given. This modelling framework is particularly well suited for reactive control approaches, which, in the case of mobile manipulation missions, are often necessary to handle the complexity of the tasks to be fulfilled, the dynamic aspect of the extended workspace and the uncertainties on the knowledge of the environment. A flexible reactive framework is thus also provided, allowing the sequencing of operational tasks (in our case, tasks described in the end-effector frame) whose natures are different but also an on-line switching mechanism between constraints that are to be satisfied using the system redundancy. This framework has been successfully implemented in simulation and on a real robot. Some of the obtained results are presented.

Object Learning Through Active Exploration

This paper addresses the problem of active object learning by a humanoid child-like robot, using a developmental approach. We propose a cognitive architecture where the visual representation of the objects is built incrementally through active exploration. We present the design guidelines of the cognitive architecture, its main functionalities, and we outline the cognitive process of the robot by showing how it learns to recognize objects in a human-robot interaction scenario inspired by social parenting. The robot actively explores the objects through manipulation, driven by a combination of social guidance and intrinsic motivation. Besides the robotics and engineering achievements, our experiments replicate some observations about the coupling of vision and manipulation in infants, particularly how they focus on the most informative objects. We discuss the further benefits of our architecture, particularly how it can be improved and used to ground concepts.

Evolutionary Robotics: Exploring New Horizons

This paper considers the field of Evolutionary Robotics (ER) from the perspective of its potential users: roboticists. The core hypothesis motivating this field of research is discussed, as well as the potential use of ER in a robot design process. Four main aspects of ER are presented: (a) ER as an automatic parameter tuning procedure, which is the most mature application and is used to solve real robotics problem, (b) evolutionary-aided design, which may benefit the designer as an efficient tool to build robotic systems (c) ER for online adaptation, i.e. continuous adaptation to changing environment or robot features and (d) automatic synthesis, which corresponds to the automatic design of a mechatronic device and its control system. Critical issues are also presented as well as current trends and pespectives in ER. A section is devoted to a roboticist’s point of view and the last section discusses the current status of the field and makes some suggestions to increase its maturity.

Control of redundant robots using learned models: An operational space control approach

We present an adaptive control approach combining forward kinematics model learning methods with the operational space control approach. This combination endows the robot with the ability to realize hierarchically organised learned tasks in parallel, using tasks null space projectors built upon the learned models. We illustrate the proposed method on a simulated 3 degrees of freedom planar robot. This system is used as a benchmark to compare our method to an alternative approach based on learning an extended Jacobian. We show the better versatility of the retained approach with respect to the latter.

Generalized hierarchical control

Multi-objective control systems for complex robots usually have to handle multiple prioritized tasks. Most existing hierarchical control techniques handle either strict task priorities by using null-space projectors or a sequence of quadratic programs; or non strict task priorities by using a weighting strategy. This paper proposes a novel approach to handle both strict and non-strict priorities of an arbitrary number of tasks. It can achieve multiple priority rearrangements simultaneously. A generalized projector, which makes it possible to completely project a task into the null-space of a set of tasks, while partially projecting it into the null-space of some other tasks, is developed. This projector can be used to perform priority transitions and task insertion or deletion. The control input is computed by solving one quadratic programming problem, where generalized projectors are adopted to maintain a task hierarchy, and equality or inequality constraints can be implemented. The effectiveness of this approach is demonstrated on a simulated robotic manipulator in a dynamic environment.

Tools for simulating humanoid robot dynamics: A survey based on user feedback

The number of tools for dynamics simulation has grown substantially in the last few years. Humanoid robots, in particular, make extensive use of such tools for a variety of applications, from simulating contacts to planning complex motions. It is necessary for the humanoid robotics community to have a systematic evaluation to assist in choosing which of the available tools is best for their research. This paper surveys the state of the art in dynamics simulation and reports on the analysis of an online survey about the use of dynamics simulation in the robotics research community. The major requirements for robotics researchers are better physics engines and open-source software. Despite the numerous tools, there is not a general-purpose simulator which dominates the others in terms of performance or application. However, for humanoid robotics, Gazebo emerges as the best choice among the open-source projects, while V-Rep is the preferred commercial simulator. The survey report has been instrumental for choosing Gazebo as the base for the new simulator for the iCub humanoid robot.

Learning Forward Models for the Operational Space Control of Redundant Robots

We present an adaptive control approach combining model learning methods with the operational space control approach. We learn the forward kinematics model of a robot and use standard algebraic methods to extract pseudo-inverses and projectors from it. This combination endows the robot with the ability to realize hierarchically organised learned tasks in parallel, using tasks null space projectors built upon the learned models. We illustrate the proposed method on a simulated 3 degrees of freedom planar robot. This system is used as a benchmark to compare our method to an alternative approach based on learning an inverse of the extended Jacobian. We show the better versatility of the retained approach with respect to the latter.

Emergence of humanoid walking behaviors from mixed-integer model predictive control

Balance strategies range from continuous postural adjustments to discrete changes in contacts: their simultaneous execution is required to maintain postural stability while considering the engaged walking activity. In order to compute optimal time, duration and position of footsteps along with the center of mass trajectory of a humanoid, a novel mixed-integer model of the system is presented. The introduction of this model in a predictive control problem brings the definition of a Mixed-Integer Quadratic Program, subject to linear constraints. Simulation results demonstrate the simultaneous adaptation of the gait pattern and posture of the humanoid, in a walking activity under large disturbances, to efficiently compromise between task performance and balance. In addition, a push recovery scenario displays how, using a single balance-performance ratio, distinct behaviors of the humanoid can be specified.