Laboratoire d’analyse et d’architecture des systèmes
Ouvrage (contribution) : Humanoid Robotics: A Reference, Springer, N°ISBN 978-94-007-6047-9, Juin 2018, 22p. , N° 18003
M.FAESSLER, A.FRANCHI, D.SCARAMUZZA
Revue Scientifique : IEEE Robotics and Automation Letters, Vol.3, N°2, pp.620-626, Avril 2018, DOI 10.1109/LRA.2017.2776353 , N° 17447
In this paper, we prove that the dynamical model of a quadrotor subject to linear rotor drag effects is differentially flat in its position and heading. We use this property to compute feed-forward control terms directly from a reference trajectory to be tracked. The obtained feed-forward terms are then used in a cascaded, nonlinear feedback control law that enables accurate agile flight with quadrotors. Compared to state-of-the-art control methods, which treat the rotor drag as an unknown disturbance, our method reduces the trajectory tracking error significantly. Finally, we present a method based on a gradient-free optimization to identify the rotor drag coefficients, which are required to compute the feed-forward control terms. The new theoretical results are thoroughly validated trough extensive comparative experiments.
G.ANTONELLI, E.CATALDI, FARRICHELLO, P.ROBUFFO-GIORDANO, S.CHIAVERINI, A.FRANCHI
UNICAS, INRIA Rennes, RIS
Revue Scientifique : IEEE Transactions on Control Systems Technology, Vol.26, N°1, pp.248-254, Janvier 2018, doi 10.1109/TCST.2017.2650679 , N° 17041
The paper presents an adaptive trajectory tracking control strategy for quadrotor Micro Aerial Vehicles (MAVs). The proposed approach, while maintaining the common assumption of an orientation dynamics faster than the translational one, removes the assumption of absence of external disturbances and of Geometric Center coincident with the Center of Mass. In particular, the trajectory tracking control law is made adaptive with respect to the presence of external forces and moments (e.g., due to wind) and to the uncertainty of parameters of the dynamic model, such as the position of the center of mass. A stability analysis is presented to analytically support the proposed controller, while numerical simulations are provided in order to validate its performance.
E.BESSIERE, A.RABILLARD, J.PRECIGOUT, L.ARBARET, L.JOLIVET, R.AUGIER, A.MENANT, N.MANSARD
Revue Scientifique : Tectonics, Janvier 2018, DOI: 10.1002/2017TC004801 , N° 18005
Although fundamental to the understanding of crustal dynamics in extensional setting, the relationships between the emplacement of granitic intrusions and activity of detachments still remain very elusive. Through a multi-scale approach, we here document a continuous deformation history between the monzogranitic intrusion of Naxos and the Naxos-Paros Detachment System (Cyclades, Greece). Field observations first show an early magmatic deformation followed by solid-state, ductile and then brittle deformation when approaching the detachment zone, as evidenced by the overprinting of mylonites by cataclastes and pseudotachylites. From these observations, we define six strain facies that characterize a positive strain gradient from core to rim of the Naxos monzogranite. Based on field pictures, X-ray tomography and Electron BackScatter Diffraction (EBSD) analyses along the strain gradient, we then quantify the intensity of mineralogical fabrics in 2D and 3D and better characterize the deformation mechanisms. Our measured shape variations of the strain ellipsoid corroborate the large-scale strain gradient, showing a good correlation between qualitative and quantitative studies. In addition, EBSD data indicate that dislocation creep was predominant during cooling from more than 500°C to temperature conditions of the ductile-to-brittle transition. However, 1) a weakening of quartz lattice preferred orientation with increasing strain and 2) evidence of numerous four-grain junctions in high-strain shear bands also indicate that grain boundary sliding significantly contributed to the deformation. Although the source of grain boundary sliding remains to be constrained, it provides a consistent approach to account for strain localization in Naxos.
M.MOREAUX, N.LYUBOVA, F.LERASLE, I.FERRANE
RAP, Softbank Robotics, IRIT-UPS
Manifestation avec acte : International Conference on Computer Vision Theory and Applications ( VISAPP ) 2018 du 27 janvier au 29 janvier 2018, Funchal (Portugal), Janvier 2018, 9p. , N° 18001
This work addresses the issue of image classification and localization of human actions based on visual data acquired from RGB sensors. Our approach is inspired by the success of deep learning in image classification. In this paper, we describe our method and how the concept of Global Average Pooling (GAP) applies in the context of semi-supervised class localization. We benchmark it with respect to Class Activation Mapping initiated in (Zhou et al., 2016), propose a regularization over the GAP maps to enhance the results, and study whether a combination of these two ideas can result in a better classification accuracy. The models are trained and tested on the Stanford 40 Action dataset (Yao et al., 2011) describing people performing 40 different actions such as drinking, cooking or watching TV. Compared to the aforementioned baseline, our model improves the classification accuracy by 5.3 percent points, achieves a localization accuracy of 50.3%, and drastically diminishes the computation needed to retrieve the class saliency from the base convolutional model.
Rapport LAAS N°18002, Janvier 2018, 70p.
Revue Scientifique : IEEE Robotics and Automation Letters, Vol.3, N°1, pp.281-288, Janvier 2018, DOI: 10.1109/LRA.2017.2738321 , N° 16242
This letter deals with the problem of controlling a robotic system whose joints have bounded position, velocity, and acceleration/torque. Assuming a discrete-time acceleration control, we compute tight bounds on the current joint accelerations that ensure the existence of a feasible trajectory in the future. Despite the clear practical importance of this issue, no complete and exact solution has been proposed yet, and all existing control architectures rely on hand-tuned heuristics. We also extend this methodology to torque-controlled robots, for which joint accelerations are only indirectly bounded by the torque limits. Numerical simulations are presented to validate the proposed method, which is computationally efficient and hence suitable for high-frequency control.
Manifestation avec acte : International Symposium on Robotics Research ( ISSR ) 2017 du 11 décembre au 14 décembre 2017, Puerto Varas (Chili), Décembre 2017, 18p. , N° 17270
We claim that navigation in human environments can be viewed as cooperative activity especially in constrained situations. Humans concurrently aid and comply with each other while moving in a shared space. Cooperation helps pedestrians to efficiently reach their own goals and respect conventions such as the personal space of others. To meet human comparable efficiency, a robot needs to predict the human trajectories and plan its own trajectory correspondingly in the same shared space. In this work, we present a navigation planner that is able to plan such cooperative trajectories, simultaneously enforcing the robot's kinematic constraints and avoiding other non-human dynamic obstacles. Using robust social constraints of projected time to a possible future collision, compatibility of human-robot motion direction, and proxemics, our planner is able to replicate human-like navigation behavior not only in open spaces but also in confined areas. Besides adapting the robot trajectory, the planner is also able to proactively propose co-navigation solutions by jointly computing human and robot trajectories within the same optimization framework. We demonstrate richness and performance of the cooperative planner with simulated and real world experiments on multiple interactive navigation scenarios.
Doctorat : INSA de Toulouse, 13 Décembre 2017, 115p., Président: P.PLOEGER, Rapporteurs: V.PADOIS, Examinateurs: A.DEL PRETE, Directeurs de thèse: F.LAMIRAUX , N° 17519
Amidst a lot of research in motion planning and control in concern with robotic applications, the mankind has never reached a point yet, where the robots are perfectly functional and autonomous in dynamic settings. Though it is controversial to discuss about the necessity of such robots, it is very important to address the issues that stop us from achieving such a level of autonomy. Industrial robots have evolved to be very reliable and highly productive with more than 1.5 million operational robots in a variety of industries. These robots work in static settings and they literally do what they are programmed for specific usecases, though the robots are flexible enough to be programmed for a variety of tasks. This research work makes an attempt to address these issues that separate both these settings in a profound way with special focus on uncertainties. Practical impossibilities of precise sensing abilities lead to a variety of uncertainties in scenarios where the robot is mobile or the environment is dynamic. This work focuses on developing smart strategies to improve the ability to handle uncertainties robustly in humanoid and industrial robots. First, we focus on a dynamical obstacle avoidance framework proposed for industrial robots equipped with skin sensors for reactivity. Path planning and motion control are usually formalized as separate problems in robotics. High dimensional configuration spaces, changing environment and uncertainties do not allow to plan real-time motion ahead of time requiring a controller to execute the planned trajectory. The fundamental inability to unify both these problems has led to handle the planned trajectory amidst perturbations and unforeseen obstacles using various trajectory execution and deformation mechanisms. The proposed framework uses ’Stack of Tasks’, a hierarchical controller using proximity information to avoid obstacles. Experiments are performed on a UR5 robot to check the validity of the framework and its potential use for collaborative robot applications. Second, we focus on a strategy to model inertial parameters uncertainties in a balance controller for legged robots. Model-based control has become more and more popular in the legged robots community in the last ten years. The key idea is to exploit a model of the system to compute precise motor commands that result in the desired motion. This allows to improve the quality of the motion tracking, while using lower feedback gains, leading so to higher compliance. However, the main flaw of this approach is typically its lack of robustness to modeling errors. In this paper we focus on the robustness of inverse-dynamics control to errors in the inertial parameters of the robot. We assume these parameters to be known, but only with a certain accuracy. We then propose a computationally-efficient optimization-based controller that ensures the balance of the robot despite these uncertainties. We used the proposed controller in simulation to perform different reaching tasks with the HRP-2 humanoid robot, in the presence of various modeling errors. Comparisons against a standard inverse-dynamics controller through hundreds of simulations show the superiority of the proposed controller in ensuring the robot balance.
A.DEL PRETE, S.TONNEAU, N.MANSARD
Rapport LAAS N°17448, Décembre 2017, 14p.
The ability to anticipate a fall is fundamental for any robot that has to balance. Currently, fast fall-prediction algorithms only exist for simple models, such as the Linear Inverted Pendulum Model (LIPM), whose validity breaks down in multi-contact scenarios (i.e. when contacts are not limited to a flat ground). This paper presents a fast fall-prediction algorithm based on the point-mass model, which remains valid in multi-contact scenarios. The key assumption of our algorithm is that, in order to come to a stop without changing its contacts, a robot only needs to accelerate its center of mass in the direction opposite to its velocity. This assumption allows us to predict the fall by means of a convex optimal control problem, which we solve with a fast custom algorithm (less than 10 ms of computation time). We validated the approach through extensive simulations with the humanoid robot HRP-2 in randomly-sampled scenarios. Comparisons with standard LIPM-based methods demonstrate the superiority of our algorithm in predicting the fall of the robot, when controlled with a state-of-the-art balance controller. This work lays the foundations for the solution of the challenging problem of push recovery in multi-contact scenarios.