IROS 2017 论文列表

Deep reinforcement learning has drawn much attention in robot control since it enables agents to learn control policies from very high dimensional states such as raw images. On the other hand, its dependency upon the availability of a significant quantity of training samples and its fragility in learning makes it difficult to apply for real world robot tasks. To alleviate these issues we propose Deep Dynamic Policy Programming (DDPP), which combines the sample efficiency and smooth policy updates of dynamic policy programming with the contemporary deep reinforcement learning framework. The effectiveness of the proposed method is first demonstrated in a simulation of the robot arm control problem, with comparison to Deep Q-Networks. As validation on a real robot system, DDPP also successfully learned the flipping of a handkerchief with a NEXTAGE humanoid robot using a reduced number of learning samples, whereas Deep Q-Networks failed to learn the task.

This paper explores a stochastic approach for controlling swarms of independent robots toward a target distribution in a bounded domain. The robot swarm has no central controller, and individual robots lack both communication and localization capabilities. Robots can only measure a scalar field (e.g. concentration of a chemical) from the environment and from this deduce the desired local swarm density. Based on this value, each robot follows a simple control law that causes the swarm as a whole to diffuse toward the target distribution. Using a new holonomic drive robot, we present the first confirmation of this control law with physical experiment. Despite deviations from assumptions underpinning the theory, the swarm achieves the theorized convergence to the target distribution in both simulation and experiment. In fact, simulated and experimental performance agree with one another and with our hypothesis that the error from the target distribution is inversely proportional to the square root of the number of robots. This is evidence that the algorithm is both practical and easily scalable to large swarms.

This paper presents a decentralized strategy for collision-free navigation of multiple agents. This strategy combines the Optimal Reciprocal Collision Avoidance (ORCA) algorithm and Model Predictive Control (MPC). Concretely, each agent applies the decentralized ORCA algorithm to compute the collision-avoiding velocities with respect to its neighbors. The derived velocities serve as constraints of a MPC problem whose solution provides the optimal control input that can ensure optimal motion of the agent. The states predicted from the agents' dynamic models are used in the ORCA algorithm to compute the ORCA velocity regions in future steps. This ORCA-MPC combined approach doesn't need a priori the preferred velocity of each agent in comparison to the traditional ORCA algorithm and its existing variants. Simulation results illustrate the effectiveness of the proposed method, and show that this new algorithm can reduce velocity vibrations in the traditional ORCA algorithm.

In this work, we propose a model-based and data efficient approach for reinforcement learning. The main idea of our algorithm is to combine simulated and real rollouts to efficiently find an optimal control policy. While performing rollouts on the robot, we exploit sensory data to learn a probabilistic model of the residual difference between the measured state and the state predicted by a simplified model. The simplified model can be any dynamical system, from a very accurate system to a simple, linear one. The residual difference is learned with Gaussian processes. Hence, we assume that the difference between real and simplified model is Gaussian distributed, which is less strict than assuming that the real system is Gaussian distributed. The combination of the partial model and the learned residuals is exploited to predict the real system behavior and to search for an optimal policy. Simulations and experiments show that our approach significantly reduces the number of rollouts needed to find an optimal control policy for the real system.

This paper addresses adaptive coverage of a spatial field without prior knowledge. Our application in this paper is to cover a region of the sea surface using a robotic boat, although the algorithmic approach has wider applicability. We propose an anytime planning technique for efficient data gathering using point-sampling based on non-uniform data-driven coverage. Our goal is to sense a particular region of interest in the environment and be able to reconstruct the measured spatial field. Since there are autonomous agents involved, there is a need to consider the costs involved in terms of energy consumed and time required to finish the task. An ideal map of the scalar field requires complete coverage of the region, but can be approximated by a good sparse coverage strategy along with an efficient interpolation technique. We propose to optimize the trade off between the environmental field mapping and the costs (energy consumed, time spent, and distance traveled) associated with sensing. We present an anytime algorithm for sampling the environment adaptively by following a multi-scale path to produce a variable resolution map of the spatial field. We compare our approach to a traditional exhaustive survey approach and show that we are able to effectively represent a spatial field spending minimum energy. We present results that indicate our sampling technique gathering most informative samples with least travel. We validate our approach through simulations and test the system on real robots in the open ocean.

Agile MAVs are required to operate in cluttered, unstructured environments at high speeds and low altitudes for efficient data gathering. Given the payload constraints and long range sensing requirements, cameras are the preferred sensing modality for MAVs. The computation burden of using cameras for obstacle sensing has forced the state of the art methods to construct world representations on a per frame basis, leading to myopic decision making. In this paper we propose a long range perception and planning approach using cameras. By utilizing FPGA hardware for disparity calculation and image space to represent obstacles, our approach and system design allows for construction of long term world representation whilst accounting for highly non-linear noise models in real time. We demonstrate these obstacle avoidance capabilities on a quadrotor flying through dense foliage at speeds of up to 4 m/s for a total of 1.6 hours of autonomous flights. The presented approach enables high speed navigation at low altitudes for MAVs for terrestrial scouting.

Robots are becoming increasingly prevalent in our society in forms where they are assisting or interacting with humans in a variety of environments, and thus they must have the ability to sense and detect objects by touch. An ongoing challenge for soft robots has been incorporating flexible sensors that can recognize complex motions and close the loop for tactile sensing. We present sensor skins that enable haptic object visualization when integrated on a soft robotic gripper that can twist an object. First, we investigate how the design of the actuator modules impact bend angle and motion. Each soft finger is molded using a silicone elastomer, and consists of three pneumatic chambers which can be inflated independently to achieve a range of complex motions. Three fingers are combined to form a soft robotic gripper. Then, we manufacture and attach modular, flexible sensory skins on each finger to measure deformation and contact. These sensor measurements are used in conjunction with an analytical model to construct 2D and 3D tactile object models. Our results are a step towards soft robot grippers capable of a complex range of motions and proprioception, which will help future robots better understand the environments with which they interact, and has the potential to increase physical safety in human-robot interaction. Please see the accompanying video for additional details.

Sub-seafloor exploration is an important area of biological and geological research, and often requires samples of seafloor mud to be recovered. The minerals and marine sediments trapped in this mud are also potentially valuable resources. However, seabed exploration has been limited by the complexity and cost of the traditional equipment used. Wide area exploration can be made possible by deploying multiple robots to autonomously search beneath the ocean floor. In this study, we prototyped a sub-seafloor excavation robot for use in seabed exploration. The key system comprised a propulsion unit and an excavation unit that allow the robot to move freely within the seabed. We experimentally demonstrated the ability of the excavation robot to produce curved boreholes with a diameter of 160 mm, a turning radius of 1.67 m, and a depth of 613 mm. Our experimental results confirm that nonlinear excavation is possible using the excavation robot developed.

We present an online system to perceive kinematic properties of articulated objects from multi-modal sensor streams. The novelty of our system is that it leverages multi-modal information in a cross-modal manner: instead of simply fusing information from different modalities, sensor streams are interpreted by leveraging information from another modality. We realize each cross-modal information extraction process using recursive estimation, with each process addressing a perceptual subproblem. Several estimators are then coupled in a cross-modal network, leading to efficient and robust online perception. We demonstrate experimentally that our cross-modal system improves over its uni-modal counterparts, increasing the variability of environments and task conditions in which the robot can robustly perceive the articulated objects. We further demonstrate that these perceptual abilities are sufficiently fast to provide feedback during manipulation actions and sufficiently comprehensive to allow the generation of new manipulation actions.

We compare five policies for teams of mobile robots to share a limited number of charging points while exploring a structured, unknown environment. Since it is infeasible to precompute an optimized schedule due to a limited time horizon, an energy-aware planner is used for adaptively deciding when and where to recharge. The coordination between robots is based on market economy giving them the flexibility to incorporate multiple objectives into a single cost function. The presented policies are evaluated through simulations with different robot team sizes on different maps. We measure the exploration gain in explored area and the exploration cost in traveled distance over time. We compare the market-based approach to a simple first-come, first-served approach. Results demonstrate the applicability for multi-robot exploration and show the strengths and weaknesses of each policy.

We present a differential geometric approach towards the synthesis of cooperative controllers for a team of autonomous surface vehicles transporting a buoyant load. We are interested in cooperative transport of large objects by teams of autonomous surface vehicles (ASVs) operating in marine and littoral environments. We consider the cooperative towing problem where individual ASVs connected to a load via cables must coordinate to transport the load along a desired trajectory. We present a differential geometric approach towards the synthesis of open and closed loop strategies for the team. The main advantage of the proposed strategy is the ability to synthesize agent-level controllers that can simultaneously satisfy all the holonomic and non-holonomic constraints within the system. We validate the approach in both simulations and experiments.

The expanding 3C (Computer, Communication, and Consumer electronics) manufacturing industry leads to a high demand on the soldering of flexible PCBs. Current manual soldering has the disadvantages of low output, low speed, and low efficiency, and upgrading soldering operations with robotic technologies is mainly limited by the property of deformation of flexible PCBs. In this paper, a novel robotic manipulation system is developed for automatic soldering of flexible PCBs, consisting of the hardware of a dual-arm configuration and the software of a cooperative control scheme. The proposed system works in a sequential manner, in the sense that a Cartesian-space region reaching controller drives an assistive arm to actively contact the PCB first, and a vision-based tracking controller activates a soldering arm after the deformation is stabilized. The proposed formulation eliminates uncertain deformation of flexible PCBs and thus guarantees the feasibility of robotic soldering.

In this article, we propose a planning algorithm for coverage of complex structures with a network of robotic sensing agents, with multi-robot surveillance missions as our main motivating application. The sensors are deployed to monitor the external surface of a 3D structure. The algorithm controls the motion of each sensor so that a measure of the collective coverage attained by the network is nondecreasing, while the sensors converge to an equilibrium configuration. A modified version of the algorithm is also provided to introduce collision avoidance properties. The effectiveness of the algorithm is demonstrated in a simulation and validated experimentally by executing the planned paths on an aerial robot.

In this paper, we apply lossless and successive convexification techniques and employ real-time on-board convex optimization to perform constrained motion planning for quadrotors. In general, this problem is challenging for real-time on-board applications due to its non-convex nature. The ability to generate feasible trajectories quickly and reliably is central to operating high-performance aerial robots in populated spaces. Motivated by our earlier research on convexification of non-convex optimal control problems, we use these convexification techniques to cast the problems into one (or a sequence of) Second-Order Cone Programming (SOCP) problem(s). In doing so, we are able to attain our solutions by leveraging modern advances in Interior Point Method (IPM) algorithms. Here, we focus on 3-degree-of-freedom trajectory generation, whereby closed-loop control is used to track a translational trajectory computed on-board at the onset of the maneuver. To the best of our knowledge, this is the first demonstration of convexification techniques used in real-time on-board trajectory generation for high-performance quad-rotor flight. We present two example scenarios: (1) a case where lossless convexification is used to increase the available control envelope, thus enabling an agile flip maneuver, and (2) a case where both convexification techniques are used to compute a path through a flight space containing ellipsoidal keep-out zones. We present flight demonstration results obtained using the Autonomous Control Laboratory's (ACL's) custom quad-rotor platforms and SOCP optimization software. Additionally, computation timing statistics for the example scenarios obtained using a series of mobile ARM and Intel processors show a minimum mean computation time of 36.5 and 122.2 milliseconds, respectively.

In this paper, we introduce a control strategy aimed at generating a stable walking pattern for a humanoid pushing a heavy load on a cart. In contrast to previous approaches that rely on force/torque sensors to measure the interaction between the robot and the pushed object, we present a simple model-based controller that can be implemented on most robots due to its computationally efficient design. Every aspect of the controller design is covered, from the formulation and validation of the dynamic model, to the implementation and validation on an actual humanoid robot. The experimental results show that the controller can efficiently make a NAO humanoid transport, in a stable way, the equivalent of its own weight on a rolling cart.

Toward stable gait generation on a steep slope, this paper discusses a method for generating a stable walking gait completing in one step for an underactuated rimless wheel. First, we introduce a model of a six-legged rimless wheel with a reaction wheel, and develop two controllers for the single-limb and double-limb support phases. For easily achieving precise control of the state variables, we apply a discrete-time output deadbeat control for each phase. We numerically analyze the fundamental properties of the generated gait, and show that the target period of the double-limb support phase must be chosen as a significantly long so that the rear and fore feet are grounded during control of the reaction wheel. The gait efficiency is also analyzed in terms of the walking speed and specific resistance through numerical simulations.

The paper presents a control method to catch a moving object with a joint actuated by means of a variable stiffness actuator. The controller is designed such that the variable stiffness joint acts as a virtual damper that absorbs the kinetic energy of the moving object. The virtual damping and the output stiffness of the variable stiffness actuator are the control variables. To obtain a critically damped system, the damping coefficient is scheduled on both the output stiffness and the inertia of the system. Experiments on the rotational variable stiffness actuator vsaUT-II validate the control method.

In recent years, hydrofoils have become ubiquitous and critical components of high-performance surface vehicles. Twenty-meter-long hydrofoil sailing craft are capable of reaching speeds in excess of 45 knots. Hydrofoil dinghies routinely travel faster than the wind and reach speeds up to 30 knots. Besides, in the quest for super-maneuverability, actuated hydrofoils could enable the efficient generation of large forces on demand. However, the control of hydrofoil systems remains challenging, especially in rough seas. With the intent to ultimately enable the design of versatile, small-scale, high-speed, and super-maneuverable surface vehicles, we investigate the problem of controlling the lift force generated by a flexible, surface-piercing hydrofoil traveling at high speed through a random wave field. We present a test platform composed of a rudder-like vertical hydrofoil actuated in pitch. The system is instrumented with velocity, force, and immersion depth sensors. We carry out high-speed field experiments in the presence of naturally occurring waves. The 2 cm chord hydrofoil is successfully controlled with a LTV/feedback linearization controller at speeds ranging from 4 to 10+ m/s.

In this study, we propose a control strategy for a power-assisted cart based on its motion state. The power-assisted cart we developed has one motor, a differential gear, and brakes. This cart uses the motor and the differential gear for moving forward, and applying brakes to either wheel allows the cart to turn both left and right. Therefore, the power-assisted cart can support the user when going straight and turning despite having only one motor. In the past we developed a control method that allows to control the cart's speed around the operation point in order to keep its magnitude constant when the cart starts turning. This was necessary, as the differential gear causes a speed change during turning, because of its characteristics. However, the desired behavior when transitioning from straight motion to turning motion is different to the desired behavior when going from turning motion to straight motion. Therefore, in this paper we propose a control method to adjust the speed in the direction of motion based on the state of the cart. We validated the effectiveness of the proposed method through experiments and discussed the results.