Dexterous in-hand manipulation is a unique and valuable human skill requiring sophisticated sensorimotor interaction with the environment while respecting stability constraints. Satisfying these constraints with generated motions is essential for a robotic platform to achieve reliable in-hand manipulation skills. Explicitly modelling these constraints can be challenging, but they can be implicitly modelled and learned through experience or human demonstrations. We propose a learning and control approach based on dictionaries of motion primitives generated from human demonstrations. To achieve this, we defined an optimization process that combines motion primitives to generate robot fingertip trajectories for moving an object from an initial to a desired final pose. Based on our experiments, our approach allows a robotic hand to handle objects like humans, adhering to stability constraints without requiring explicit formalization. In other words, the proposed motion primitive dictionaries learn and implicitly embed the constraints crucial to the in-hand manipulation task.

Dexterous in-hand manipulation is a peculiar and useful human skill. This ability requires the coordination of many senses and hand motion to adhere to many constraints. These constraints vary and can be influenced by the object characteristics or the specific application. One of the key elements for a robotic platform to implement reliable in-hand manipulation skills is to be able to integrate those constraints in their motion generations. These constraints can be implicitly modelled, learned through experience or human demonstrations. We propose a method based on motion primitives dictionaries to learn and reproduce in-hand manipulation skills. In particular, we focused on fingertip motions during the manipulation, and we defined an optimization process to combine motion primitives to reach specific fingertip configurations. The results of this work show that the proposed approach can generate manipulation motion coherent with the human one and that manipulation constraints are inherited even without an explicit formalization.

Tactile object shape identification is important for robotic hands to perform dexterous manipulation. Most of the proposed approaches concentrate on specific object recognition. This limits their application into more realistic environments where a larger amount of objects are present. We present a method that performs object shape identification independently on the size and location of the object within the hand. This method allows sequential learning of new shapes. The method combines proprioceptive signatures and contact normal information to build a descriptor that reduces the impact of the size and pose of the object on recognition rate. Sequential training is performed with a collection of Neural Forests (NF). This allows adding new objects to the training set so that training the model from scratch is avoided. Extensive experiments reveal that the combination of multiple modalities (e.g. contact normals, proprioceptive information) is beneficial to the system accuracy, and that results for sequential learning are on par with its batch counterpart. This makes sequential training advantageous because it is less time consuming. Results showed that both techniques depict similar results and perform with at least 83% in a 7-shapes case scenario in a simulated environment. Experiments are made with a real shadow hand using a NF trained on simulated data.

Most modern approaches for tactile object recognition with robotic hands do not use proprioceptive data. In those that do, a limited number of objects with similar shapes is recognized. Furthermore, Self-Organizing Maps (SOM) based on raw values of joint angles/torques are frequently implemented which requires large sets of training data. In this paper, we present an approach based only on joint angles of a robotic hands to identify the shape of an object regardless its size and position within the hand. A representation of the joint angles is created to endow the robotic hand with proprioception. Support Vector Machine (SVM) is implemented for shape identification using patterns or signatures generated on this representation when the objects are grasped. To illustrate the scope of this method, tests are performed on five shapes present in common objects. Both SVM and SOM trained with signatures were at least 10% more accuracy than the ones trained with raw values of joint angles. Training sets are reduced at least 85% with respect to other works. An accuracy of 94% was obtained on large ranges of dimensions of the shapes and positions.

In the process of developing human-like robotic hands, engineers are looking for robust and accurate sensors that can fit in very confined spaces. Rotary encoders that measure joint angles of a multi-fingered hand must be of very small size and yet provide reliable and repetitive data. Hall-effect sensors combined with ring magnets fit in narrow spaces and have proved to be durable rotary magnetic position sensors due to their contact-less features. However, when several sensors of this kind are packed closely together in a finger, magnetic crosstalk effects appear and can lead to important errors/shifts in nearby sensor readings. This paper describes magnetic crosstalk effects in nearby joints and proposes a compensation method directly linked with the software calibration process.

This paper presents a novel planner for dexterous manipulation with a multi-fingered robotic hand. This planner receives as input an initial grasp of the object and a desired trajectory of the object in task space (position and orientation). The planner computes the movements of the fingers which are required to move the object along this trajectory without breaking contacts. It uses a triangle mesh representation of the surfaces of the fingers and the object and a new contact evolution graph in order to compute all the possible transitions between the contact primitives of their surfaces. This planner have been implemented as a program which communicates with a five-fingered hand in order to test them in real manipulation tasks.

This work deals with grasping using an anthropomorphic hand. The main idea is to easily compute a grasp for a robotic hand in the context of a given task. This paper describes a method that does not require learning. Starting from works in the neuroscience field on human hand postural synergies, we introduce a two-level algorithm that uses a mathematical model of relationships between muscles and degrees-of-freedom of the hand and a set of five parameters to define synergies between muscles according to some grasp properties taken from an existing taxonomy of grasps. The two-level architecture presented in this paper aims to provide the flexibility needed for working with a real robotic hand. This algorithm is validated both in simulation using Gazebo and on the Shadow Robot Hand.

Despite recent progress, the performance of force control algorithms still appears to be poor when applying to systems with significant backlash, low precision of position sensors, low communication bandwidth and computation power. Anthropomorphic robot hands with tendon driven joints are typical examples of such systems. To overcome this difficulty, this paper proposes an approach that uses the torque saturation (max-torque) of the joint position control loops to control the end-effector (fingertip) force. This control scheme has been implemented and tested on the Shadow motor robot hand. An application of this control scheme has also been implemented for two fingers holding and rotating an object around the vertical axis. This experiment shows the strong potential of this force control algorithm for grasping and dexterous manipulation activities.

We propose a method to plan in-hand manipulation actions with a robotic anthropomorphic hand. We consider in-hand manipulation actions as sequences between canonical grasp types identified in the humans. Our work concerns the generation of this sequence, which should be autonomous and fast enough to be performed on-line. We use a Markov Decision Process (MDP) governing the transitions between grasp types, depending on the object and on the goal grasp. The policy is learnt directly from human behavior, after an initialization using an empirical estimation of the state action probabilities of the MDP. Then, the policy is finely learnt from samples of human in-hand manipulation records. These samples are chosen using active learning, in order to maximize the useful information of every record, and speed up the learning process. For planning, the policy gives the sequence with highest probability of success. We show a serie of realistic human-like grasp transition sequences derived from the proposed method.

For an anthropomorphic arm-hand robot, grasping and in-hand manipulating an object can be realized with numerous approach trajectories and grasping configurations. The redundancy at this level of the tasks is due to a large number of degrees-of-freedom (DOFs) of the arm-hand system. This redundancy constitutes a big challenge to the planning and control tasks of the robot. For this kind of tasks, human has his own choices privileging certain configurations over the others. These choices come from a long learning process which implicitly takes into account the mechanical constraints of the system. In this work, we concentrate our effort on deciphering certain mechanical constraints, ¿tenodesis¿ phenomenon in particular, in order to solve the redundancy and imitate the human natural gestures in the tasks of grasping or in-hand manipulation.

We propose in this paper a new concept of unified position/orientation control of robot manipulator by describing the end-effector motion as a dual quaternion involving both translation and rotation. The development of the forward kinematic model and Jacobian matrix in dual quaternion space is detailed as well as the stability of the controller. At last, simulation and experimental results highlight the efficiency and performance of this controller.

A grasp is the beginning of any manipulation task. Therefore, an autonomous robot should be able to grasp objects it sees for the first time. It must hold objects appropriately in order to successfully perform the task. This paper considers the problem of grasping unknown objects in the same manner as humans. Based on the idea that the human brain represents objects as volumetric primitives in order to recognize them, the presented algorithm predicts grasp as a function of the object's parts assembly. Beginning with a complete 3D model of the object, a segmentation step decomposes it into single parts. Each single part is fitted with a simple geometric model. A learning step is finally needed in order to find the object component that humans choose to grasp it.

In this paper, we propose a new method for the motion planning problem of rigid object dexterous manipulation with a robotic multi-fingered hand, under quasi-static movement assumption. This method computes both object and finger trajectories as well as the finger relocation sequence. Its specificity is to use a special structuring of the research space that allows to search for paths directly in the particular subspace GSnwhich is the subspace of all the grasps that can be achieved with n grasping fingers. The solving of the dexterous manipulation planning problem is based upon the exploration of this subspace. The proposed approach captures the connectivity of GSnin a graph structure. The answer of the manipulation planning query is then given by searching a path in the computed graph. Simulation experiments were conducted for different dexterous manipulation task examples to validate the proposed method.

This paper addresses the motion planning problem of the dexterous manipulation of 3D rigid objects by a robotic multi-fingered hand. We propose a novel approach based on probabilistic roadmap techniques. Inspired by the theory developed by Alami et al. (1994), Simeon et al. (2003), the planner relies on a topological property that characterizes the existence of solutions in GSn, a specific manifold of the configuration space. This property leads to reduce the problem by structuring the search-space. It allows us to design a manipulation planner that directly captures in a probabilistic roadmap the connectivity of sub-dimensional manifolds of the composite configuration space. The proposed method allows a global planning - both object and fingers trajectories are computed - that can cope with the obstacle presence in the environment. Collisions between different fingers or between object and fingers elsewhere than fingertips are avoided. Force closure constraints are taken into account to ensure the computed paths physical feasibility, under quasi-static motion assumption. First experiments demonstrate the feasibility and the efficiency of the approach

This paper presents a new solution for solving the grasping force optimization problem, fundamental in dexterous manipulation by multifingered robotic hands. Several methods have been proposed in the literature, yielding optimal solutions, with either recursive or non linear programming techniques. However, most of them involve many computations and cannot be used online. Furthermore, they do not offer a smooth solution regarding to possible changes in the contact conditions due to finger rolling or gaiting, or in the desired resultant force to be exerted on the grasped object. The more recent ones are fast and smooth enough for real-time computation but the method we present here is faster, easier to implement and provide very satisfying results, even though the solution is sub-optimal. The method is based on the minimization of a cost function that gives an analytical solution but does not ensure by itself the satisfaction of the static frictional constraints. An associated iterative adjustment modifies this function until the internal forces enter the friction cone. The minimal solution is found within a few iterations. Force determination is therefore included in the simulation of a hybrid position/force controller to prove the effectiveness of such an approach for updating the force references during the grasped object motion.

We deal with the control of a multifingered robot hand in a dexterous manipulation task. For that purpose, we have previously proposed an efficient modular and hierarchical control scheme. This solution is different from other proposed methods since the position and force controllers are decentralized, that is, implemented at the local level of each finger. A new module is introduced here to compensate for the error caused by the deviation of the contact locations due to the rolling contacts between the fingertips and the handled object. On the basis of the contact kinematics, the "rolling contact" module updates the contact locations to modify the position and force references sent to each local finger controller in order to account for the motion of the contact points.

In this paper, a penalty approach which deals with a constrained optimization to solve the inverse kinematic of redundant robot manipulators is considered. An optimization procedure using neural networks is formulated, it produces on-line position and velocity trajectories in joint space from position and orientation trajectories in Cartesian space. This new method offers substantially better accuracy and guarantees a good minimization of a performance function subject to joint limitations while achieving the end-effector task.