Search

Complex manipulation tasks can contain various execution branches of primitive skills in sequence or in parallel under different scenarios. Manual specifications of such branching conditions and associated skill parameters are not only error-prone due to corner cases, but also quickly untraceable given a large number of objects and skills. On the other hand, learning from demonstration has increasingly shown to be an intuitive and effective way to program such skills for industrial robots. Parameterized skill representations allow generalization over new scenarios, which however makes the planning process much slower thus unsuitable for online applications. In this article, we propose a hierarchical and compositional planning framework that learns a geometric task network (GTN) from exhaustive planners, without any manual inputs. A GTN is a goal-dependent task graph that encapsulates both the transition relations among skill representations and the geometric constraints underlying these transitions. This framework has shown to improve dramatically the offline learning efficiency, the online performance, and the transparency of decision process, by leveraging the task-parameterized models. We demonstrate the approach on a 7-DoF robot arm both in simulation and on hardware solving various manipulation tasks. [ABSTRACT FROM AUTHOR]

In robotics, ergodic control extends the tracking principle by specifying a probability distribution over an area to cover instead of a trajectory to track. The original problem is formulated as a spectral multiscale coverage problem, typically requiring the spatial distribution to be decomposed as Fourier series. This approach does not scale well to control problems requiring exploration in search space of more than two dimensions. To address this issue, we propose the use of tensor trains, a recent low-rank tensor decomposition technique from the field of multilinear algebra. The proposed solution is efficient, both computationally and storagewise, hence making it suitable for its online implementation in robotic systems. The approach is applied to a peg-in-hole insertion task requiring full 6-D end-effector poses, implemented with a seven-axis Franka Emika Panda robot. In this experiment, ergodic exploration allows the task to be achieved without requiring the use of force/torque sensors. [ABSTRACT FROM AUTHOR]

This article considers the problem of planning a trajectory that maximizes the probability that a robot will be able to complete a set of point-to-point tasks, after experiencing locked joint failures. The proposed approach first develops a method to calculate the probability of task failure for an arbitrary trajectory based on its failure scenarios, which are efficiently computed by identifying the ranges of task point self-motion manifolds. Then, a novel trajectory planning algorithm is proposed to find the optimal trajectory with maximum probability of task completion. The planning algorithm exploits the overlap of self-motion manifold bounding boxes, as opposed to always using the shortest distance, to determine an optimal trajectory. The proposed trajectory planning algorithm is demonstrated on planar positioning 3R, spatial positioning 4R, and spatial positioning/orienting 7R redundant robots, resulting in average improvement of 17%, 22%, and 30%, respectively, compared to the best shortest distance trajectory. [ABSTRACT FROM AUTHOR]

Designing artificial avatars able to interact with humans in a safe, smart, and natural way is a current open problem in control. Solving such an issue will allow the design of cyber-agents capable of cooperatively interacting with people in order to fulfil common joint tasks in a multitude of different applications. This is particularly relevant in the context of healthcare applications. Indeed, the use for rehabilitation has been proposed of artificial agents able to interact and coordinate their movements with those of patients suffering from social or motor disorders. Moreover, it has also been shown that the level of motor coordination between the avatar and the human patient is enhanced if the kinematic properties of the avatar’s motion are similar to those of the individual it is interacting with. In this article, we discuss, first, a new method based on Markov chains to confer “human motor characteristics” on the motion of a virtual agent so that it can coordinate its motion with that of a target individual while exhibiting specific kinematic properties. Then, we embed such synthetic model in a novel control architecture based on reinforcement learning to synthesize a cyber-agent able to mimic the behavior of a specific human performing a joint motor task with one or more individuals. [ABSTRACT FROM AUTHOR]

Multirobot fleets play an important role in industrial logistics, surveillance, and exploration applications. A wide literature exists on the topic, both resorting to reactive (i.e. collision avoidance) and to deliberative (i.e. motion planning) techniques. In this work, null space-based inverse kinematics (NSB-IK) methods are applied to the problem of fleet management. Several NSB-IK approaches existing in the literature are reviewed, and compared with a reverse priority approach, which originated in manipulator control, and is here applied for the first time to the considered problem. All NSB-IK approaches are here described in a unified formalism, which allows (i) to encode the property of each controller into a set of seven main key features, (ii) to study possible new control laws with an opportune choice of these parameters. Furthermore, motivated by the envisioned application scenario, we tackle the problem of task-switching activation. Leveraging on the iCAT TPC technique Simetti and Casalino, 2016, in this article, we propose a method to obtain continuity in the control in face of activation or deactivation of tasks, and subtasks by defining suitable damped projection operators. The proposed approaches are evaluated formally, and via simulations. Performances with respect to standard methods are compared considering a specific case study for multivehicles management. [ABSTRACT FROM AUTHOR]

The task of shape formation in robot swarms can often be reduced to two tasks—assigning goal locations to each robot and creating a collision-free path to that goal. In this article, we present a distributed algorithm that solves these tasks concurrently, enabling a swarm of robots to move and form a shape quickly and without collision. A user can specify a desired shape as an image, send that to a swarm of identically programmed robots, and the swarm will move all robots to goal locations within the desired shape. This algorithm was executed on a swarm of up to 1024 simulated robots and a swarm of 100 real robots, showing that it reliably converges to all robots forming the shape. [ABSTRACT FROM AUTHOR]

In this article, a passive physical human–robot interaction (pHRI) controller is proposed to enhance pHRI performance in terms of precision, cognitive load, and user effort, in cases partial knowledge of the task is available. Partial knowledge refers to a subspace of $SE(3)$ determined by the desired task, generally mixing both position and orientation variables, and is mathematically approximated by parametric expressions. The proposed scheme, which utilizes the notion of virtual constraints and the prescribed performance control methodology, is proved to be passive with respect to the interaction force, while guaranteeing constraint satisfaction in all cases. The control scheme is experimentally validated and compared with a dissipative control scheme utilizing a KUKA LWR4+ robot in a master–slave task; experiments also include an application to a robotic assembly case. [ABSTRACT FROM AUTHOR]

Contact-rich manipulation tasks in unstructured environments often require both haptic and visual feedback. It is nontrivial to manually design a robot controller that combines these modalities, which have very different characteristics. While deep reinforcement learning has shown success in learning control policies for high-dimensional inputs, these algorithms are generally intractable to train directly on real robots due to sample complexity. In this article, we use self-supervision to learn a compact and multimodal representation of our sensory inputs, which can then be used to improve the sample efficiency of our policy learning. Evaluating our method on a peg insertion task, we show that it generalizes over varying geometries, configurations, and clearances, while being robust to external perturbations. We also systematically study different self-supervised learning objectives and representation learning architectures. Results are presented in simulation and on a physical robot. [ABSTRACT FROM AUTHOR]

In this article, we study multirobot task allocation problems where task costs vary. The variation may be, for example, due to the revelation of new information or other dynamic circumstances. As robots update their cost estimates, typically they will update task assignments to reflect the new information using additional communication and computation. In dynamic settings, the robots are continually repairing the optimality of the system's task assignments, which can incur substantial communication and computation. We investigate how one can reduce communication and centralized computation expense during execution by using a prior model of how costs may change and performing upfront computation of possible robot–task assignments. First, we develop an algorithm that partitions a team of robots into several independent subteams that are able to maintain global optimality by communicating entirely amongst themselves. Second, we propose a method for computing the worst-case cost suboptimality if robots persist with the initial assignment and perform no further communication and computation. Finally, we introduce an algorithm to assess whether cost changes affect the optimality of the current assignment through a succession of local communication exchanges. Experimental results show that the proposed methods are helpful in reducing the degree of centralization needed by a multirobot system (e.g., the third method gave at least 45% reduction of global communication across all scenarios studied). The methods are valuable in transitioning multirobot techniques, which have met with success in structured applications (such as factories and warehouses) to the broader, wilder world. [ABSTRACT FROM AUTHOR]

This article considers a multiplayer reach–avoid game between two adversarial teams in a general convex domain which consists of a target region and a play region. The evasion team, initially lying in the play region, aims to send as many team members into the target region as possible, while the pursuit team with its team members initially distributed in both play region and target region, strives to prevent that by capturing the evaders. We aim at investigating a task assignment about the pursuer–evader matching, which can maximize the number of the evaders who can be captured before reaching the target region safely when both teams play optimally. To address this, two winning regions for a group of pursuers to intercept an evader are determined by constructing an analytical barrier which divides these two parts. Then, a task assignment to guarantee the most evaders intercepted is provided by solving a simplified 0-1 integer programming instead of a nondeterministic polynomial problem, easing the computation burden dramatically. It is worth noting that except the task assignment, the whole analysis is analytical. Finally, simulation results are also presented. [ABSTRACT FROM AUTHOR]