Deep Reinforcement Learning for Attacking Wireless Sensor Networks.

Recent advances in Deep Reinforcement Learning allow solving increasingly complex problems. In this work, we show how current defense mechanisms in Wireless Sensor Networks are vulnerable to attacks that use these advances. We use a Deep Reinforcement Learning attacker architecture that allows having one or more attacking agents that can learn to attack using only partial observations. Then, we subject our architecture to a test-bench consisting of two defense mechanisms against a distributed spectrum sensing attack and a backoff attack. Our simulations show that our attacker learns to exploit these systems without having a priori information about the defense mechanism used nor its concrete parameters. Since our attacker requires minimal hyper-parameter tuning, scales with the number of attackers, and learns only by interacting with the defense mechanism, it poses a significant threat to current defense procedures.

Autonomous Robots for Space: Trajectory Learning and Adaptation Using Imitation.

This paper adds on to the on-going efforts to provide more autonomy to space robots and introduces the concept of programming by demonstration or imitation learning for trajectory planning of manipulators on free-floating spacecraft. A redundant 7-DoF robotic arm is mounted on small spacecraft dedicated for debris removal, on-orbit servicing and assembly, autonomous and rendezvous docking. The motion of robot (or manipulator) arm induces reaction forces on the spacecraft and hence its attitude changes prompting the Attitude Determination and Control System (ADCS) to take large corrective action. The method introduced here is capable of finding the trajectory that minimizes the attitudinal changes thereby reducing the load on ADCS. One of the critical elements in spacecraft trajectory planning and control is the power consumption. The approach introduced in this work carry out trajectory learning offline by collecting data from demonstrations and encoding it as a probabilistic distribution of trajectories. The learned trajectory distribution can be used for planning in previously unseen situations by conditioning the probabilistic distribution. Hence almost no power is required for computations after deployment. Sampling from a conditioned distribution provides several possible trajectories from the same start to goal state. To determine the trajectory that minimizes attitudinal changes, a cost term is defined and the trajectory which minimizes this cost is considered the optimal one.

Haptic-Guided Teleoperation of a 7-DoF Collaborative Robot Arm With an Identical Twin Master.

In this article, we describe two techniques to enable haptic-guided teleoperation using 7-DoF cobot arms as master and slave devices. A shortcoming of using cobots as master-slave systems is the lack of force feedback at the master side. However, recent developments in cobot technologies have brought in affordable, flexible, and safe torque-controlled robot arms, which can be programmed to generate force feedback to mimic the operation of a haptic device. In this article, we use two Franka Emika Panda robot arms as a twin master-slave system to enable haptic-guided teleoperation. We propose a two layer mechanism to implement force feedback due to 1) object interactions in the slave workspace, and 2) virtual forces, e.g. those that can repel from static obstacles in the remote environment or provide task-related guidance forces. We present two different approaches for force rendering and conduct an experimental study to evaluate the performance and usability of these approaches in comparison to teleoperation without haptic guidance. Our results indicate that the proposed joint torque coupling method for rendering task forces improves energy requirements during haptic guided telemanipulation, providing realistic force feedback by accurately matching the slave torque readings at the master side.

A Haptic Shared-Control Architecture for Guided Multi-Target Robotic Grasping.

Although robotic telemanipulation has always been a key technology for the nuclear industry, little advancement has been seen over the last decades. Despite complex remote handling requirements, simple mechanically linked master-slave manipulators still dominate the field. Nonetheless, there is a pressing need for more effective robotic solutions able to significantly speed up the decommissioning of legacy radioactive waste. This paper describes a novel haptic shared-control approach for assisting a human operator in the sort and segregation of different objects in a cluttered and unknown environment. A three-dimensional scan of the scene is used to generate a set of potential grasp candidates on the objects at hand. These grasp candidates are then used to generate guiding haptic cues, which assist the operator in approaching and grasping the objects. The haptic feedback is designed to be smooth and continuous as the user switches from a grasp candidate to the next one, or from one object to another one, avoiding any discontinuity or abrupt changes. To validate our approach, we carried out two human-subject studies, enrolling 15 participants. We registered an average improvement of 20.8%, 20.1%, and 32.5% in terms of completion time, linear trajectory, and perceived effectiveness, respectively, between the proposed approach and standard teleoperation.

Directly Printable Flexible Strain Sensors for Bending and Contact Feedback of Soft Actuators.

This paper presents a fully printable sensorized bending actuator that can be calibrated to provide reliable bending feedback and simple contact detection. A soft bending actuator following a pleated morphology, as well as a flexible resistive strain sensor, were directly 3D printed using easily accessible FDM printer hardware with a dual-extrusion tool head. The flexible sensor was directly welded to the bending actuator's body and systematically tested to characterize and evaluate its response under variable input pressure. A signal conditioning circuit was developed to enhance the quality of the sensory feedback, and flexible conductive threads were used for wiring. The sensorized actuator's response was then calibrated using a vision system to convert the sensory readings to real bending angle values. The empirical relationship was derived using linear regression and validated at untrained input conditions to evaluate its accuracy. Furthermore, the sensorized actuator was tested in a constrained setup that prevents bending, to evaluate the potential of using the same sensor for simple contact detection by comparing the constrained and free-bending responses at the same input pressures. The results of this work demonstrated how a dual-extrusion FDM printing process can be tuned to directly print highly customizable flexible strain sensors that were able to provide reliable bending feedback and basic contact detection. The addition of such sensing capability to bending actuators enhances their functionality and reliability for applications such as controlled soft grasping, flexible wearables, and haptic devices.

Learning modular policies for robotics.

A promising idea for scaling robot learning to more complex tasks is to use elemental behaviors as building blocks to compose more complex behavior. Ideally, such building blocks are used in combination with a learning algorithm that is able to learn to select, adapt, sequence and co-activate the building blocks. While there has been a lot of work on approaches that support one of these requirements, no learning algorithm exists that unifies all these properties in one framework. In this paper we present our work on a unified approach for learning such a modular control architecture. We introduce new policy search algorithms that are based on information-theoretic principles and are able to learn to select, adapt and sequence the building blocks. Furthermore, we developed a new representation for the individual building block that supports co-activation and principled ways for adapting the movement. Finally, we summarize our experiments for learning modular control architectures in simulation and with real robots.

Stochastic optimal control methods for investigating the power of morphological computation.

One key idea behind morphological computation is that many difficulties of a control problem can be absorbed by the morphology of a robot. The performance of the controlled system naturally depends on the control architecture and on the morphology of the robot. Because of this strong coupling, most of the impressive applications in morphological computation typically apply minimalistic control architectures. Ideally, adapting the morphology of the plant and optimizing the control law interact so that finally, optimal physical properties of the system and optimal control laws emerge. As a first step toward this vision, we apply optimal control methods for investigating the power of morphological computation. We use a probabilistic optimal control method to acquire control laws, given the current morphology. We show that by changing the morphology of our robot, control problems can be simplified, resulting in optimal controllers with reduced complexity and higher performance. This concept is evaluated on a compliant four-link model of a humanoid robot, which has to keep balance in the presence of external pushes.

Learned graphical models for probabilistic planning provide a new class of movement primitives.

BIOLOGICAL MOVEMENT GENERATION COMBINES THREE INTERESTING ASPECTS: its modular organization in movement primitives (MPs), its characteristics of stochastic optimality under perturbations, and its efficiency in terms of learning. A common approach to motor skill learning is to endow the primitives with dynamical systems. Here, the parameters of the primitive indirectly define the shape of a reference trajectory. We propose an alternative MP representation based on probabilistic inference in learned graphical models with new and interesting properties that complies with salient features of biological movement control. Instead of endowing the primitives with dynamical systems, we propose to endow MPs with an intrinsic probabilistic planning system, integrating the power of stochastic optimal control (SOC) methods within a MP. The parameterization of the primitive is a graphical model that represents the dynamics and intrinsic cost function such that inference in this graphical model yields the control policy. We parameterize the intrinsic cost function using task-relevant features, such as the importance of passing through certain via-points. The system dynamics as well as intrinsic cost function parameters are learned in a reinforcement learning (RL) setting. We evaluate our approach on a complex 4-link balancing task. Our experiments show that our movement representation facilitates learning significantly and leads to better generalization to new task settings without re-learning.

Biologically inspired kinematic synergies enable linear balance control of a humanoid robot.

Despite many efforts, balance control of humanoid robots in the presence of unforeseen external or internal forces has remained an unsolved problem. The difficulty of this problem is a consequence of the high dimensionality of the action space of a humanoid robot, due to its large number of degrees of freedom (joints), and of non-linearities in its kinematic chains. Biped biological organisms face similar difficulties, but have nevertheless solved this problem. Experimental data reveal that many biological organisms reduce the high dimensionality of their action space by generating movements through linear superposition of a rather small number of stereotypical combinations of simultaneous movements of many joints, to which we refer as kinematic synergies in this paper. We show that by constructing two suitable non-linear kinematic synergies for the lower part of the body of a humanoid robot, balance control can in fact be reduced to a linear control problem, at least in the case of relatively slow movements. We demonstrate for a variety of tasks that the humanoid robot HOAP-2 acquires through this approach the capability to balance dynamically against unforeseen disturbances that may arise from external forces or from manipulating unknown loads.