ICRA 2019 Papers

As robotic platforms have become more capable and autonomous, they have increasingly been utilized in time sensitive applications such as search and rescue. To that end, we have developed a system for teams of robots to efficiently explore an environment while taking sensor measurements. The system utilizes an information seeking algorithm that generates high priority points of interest based on the highest expected information gained per distance travelled. In order to coordinate multiple robots, the system partitions the area into different regions according to the effort needed to explore each region. Robots are assigned different regions to measure in order to minimize repetition of work and reduce interference between each robot.We present an information rate adaptive sampling approach for tasking robots within an environment to gather sensor measurements. We evaluated our approach within a simulation environment with one to four robots. Multiple robots are coordinated through our region segmentation approach. The data shows efficiency gains through the use of adaptive information gain rate tasking above a naïve closest point approach. We also see positive results from using the region segmentation technique. We further the experimentation by testing the algorithm on real world robots and verify the results in real world experimentation.

We consider the problem of multiple robots that must cooperate within a shared environment, but which wish to limit the information they disclose during their coordination efforts. Specifically, we examine the problems of privacy-preserving rendezvous and persistent monitoring. In the former, the robots construct a joint plan to have them meet, without either knowing beforehand where or when the meeting will occur. In the latter, multiple robots dynamically cover a region of space-they plan collective motions which are collision-free but with the assurance that agents remain ignorant of the paths of others. Accordingly, the tasks are sort of inverses in that the robots must collectively determine whether their joint paths collide or not, then, using this, achieve their collective task. Other than what is learned by the outcome of the joint-collision determination, the robots possess no details of the other paths. Our approach builds on garbled circuits and homomorphic encryption to realize basic secure path intersection primitives. We present algorithms, a software implementation, and a physical experiment on mobile robots to test the practical feasibility of our approach. We believe that these ideas provide a valuable direction for adoption in small Unmanned Systems belonging to different stakeholders.

We propose a feedback control system for a reconfigurable multi-vessel platform. The platform consists of N propeller-driven vessels each of which is capable of latching to another vessel to form a rigid body of connected vessels. The main technical challenges are that i) depending on configurations of the platform the dynamic model would be different, and ii) the number of control variables in control system design increases as does the total number of vessels in the platform. To address these challenges, we develop a coordinated robust control scheme. Through experiments we assess trajectory tracking and disturbance attenuation performance of the control scheme in various configurations of the platform. Experiment results yield that average position and orientation tracking error are approximately 0.09m and 3°, and the maximum tracking error-to-disturbance ratio is 1.12.

There is value in exploring new robot morphologies that combine the benefits of legged robots (i.e. high terrain adaptability, humanlike form factor) with those of wheeled systems (i.e. high stability, low energy consumption). Once the basic design concept has been demonstrated, there is a further requirement to validate the controllability of the system through adaptation and implementation of advanced control techniques. This research presents a motion planner that enables the Aerobot robot, an actively articulated wheeled humanoid robot capable of performing step climbing and gap crossing, to plan dynamically stable motion in response to a range of task conditions. Using a model of the robots kinematics, a motion planner is developed. The sequential quadratic program algorithm is used to solve quadratic objectives and non-linear constraints pertaining to stability and postural configuration adjustment requirements. The results showed good tracking of the robot's Zero Moment Point while transitioning between postural configurations and traversing obstacles. During these phase transitions, robustness in maintaining balance and position is also demonstrated via velocity control of the drive wheels.

We consider the feedback design for stabilizing a rigid body system by making and breaking multiple contacts with the environment without prespecifying the timing or the number of occurrence of the contacts. We model such a system as a discrete-time hybrid polynomial system, where the state-input space is partitioned into several polytopic regions with each region associated with a different polynomial dynamics equation. Based on the notion of occupation measures, we present a novel controller synthesis approach that solves finite-dimensional semidefinite programs as approximations to an infinite-dimensional linear program to stabilize the system. The optimization formulation is simple and convex, and for any fixed degree of approximations the computational complexity is polynomial in the state and control input dimensions. We illustrate our approach on some robotics examples.

Soft robot is an emergent research field which has variant promising applications. However, the design of soft robots nowadays still follows the trial-and-error process, which is not at all efficient. This paper proposes to design soft robots by pre-checking controllability during the numerical design phase. Finite-element method is used to model the dynamics of silicone soft robots, based on which the differential geometric method is applied to analyze the controllability of the points of interest. Such a verification is also investigated via model order reduction technique and Galerkin projection. The proposed methodology is finally validated by numerically designing a controllable parallel soft robot.

This paper describes development of a controllerfor bilateral robotic system for tele-echography with delayed communications. The system comprises two identical 6DOF robotic devices based on Stewart-Gough mechanisms, which were developed especially for the sake of the considered application. Controller design is facilitated by a symmetry of the teleoperation setup, in which the master and slave devices have identical geometry, actuation and measurement arrays. To further simplify the analysis, we treat coupling between degrees of freedom as exogenous disturbances and this allows us to split the control problem into six independent IDOF settings. The IDOF problems are addressed then using a novel approach to bilateral teleoperation control, based on a complete parameterization of feasible teleoperators. It allows to intuitively shape teleoperator performance while guaranteeing passivity and thus coupled stability of the system. The potential of the proposed control method is demonstrated with experimental results.

Due to agile product development, production systems have to be flexible and adaptable to meet high quality standards and a high productivity. As a result, set-up processes has to be shorter and more resilient because of the increasing number of variants. To meet future requirements, the processes need to be self-adaptive and reconfigurable at any time. Nowadays, a shift of the automation pyramid to interconnected cyber physical systems can be observed as well as emerging technologies as cloud and edge computing are introduced to production systems. These technologies in combination with the Digital Shadow, which provides information about all production assets, open up potential for an adaptive process control and an overall life cycle data management. For this, the Digital Shadow has to be used not only for the aggregation of data, but also for pushing data back into the system and to control the process. As a result, services in regard to an architecture based on edge computing as an enabling technology for an adaptive production together with the Digital Shadow are presented, implemented and discussed on the basis of an industrial use case.

This paper investigates the effect of the robot configuration on the performance of an aerial cable towed system (ACTS) composed of three quadrotors manipulating a point mass payload. The kinematic and dynamic models of the ACTS are derived in a minimal set of geometric coordinates, and a centralized feedback linearization controller is developed. Independent to the payload trajectory, the configuration of the ACTS is controlled and is evaluated using a robustness index named the capacity margin. Experiments are performed with optimal, suboptimal, and wrench infeasible configurations. It is shown that configurations near the point of zero capacity margin allow the ACTS to hover but not to follow dynamic trajectories, and that the ACTS cannot fly with a negative capacity margin. Dynamic tests of the ACTS show the effects of the configuration on the achievable accelerations.

We propose a novel method adaptive subdivision (AS) to evaluate the distance function for moving general polygonal models. The distance function can have a positive and a negative value, each of which corresponds to the Euclidean distance and penetration depth, respectively. In our approach, the distance between a pair of objects can be evaluated along any time interval of the object’s trajectory; therefore it is called “continuous”, and a minimum of the continuous distance (MCD) is determined for collision avoidance. In order to compute a MCD for general polygonal models, we calculate the upper and lower bounds of the distance in the time interval and abandons the time intervals that cannot realize the MCD. We have implemented our distance evaluation method, and have experimentally validated the proposed methods to effectively and accurately find the MCDs to generate a collision-free motion for the HRP-2 humanoid robot.

This paper presents a novel model-free Reinforcement Learning algorithm for learning behavior in continuous action, state, and goal spaces. The algorithm approximates optimal value functions using non-parametric estimators. It is able to efficiently learn to reach multiple arbitrary goals in deterministic and nondeterministic environments. To improve generalization in the goal space, we propose a novel sample augmentation technique. Using these methods, robots learn faster and overall better controllers. We benchmark the proposed algorithms using simulation and a real-world voltage controlled robot that learns to maneuver in a non-observable Cartesian task space.

Mapping the occupancy of an environment is central for robot autonomy. Traditional occupancy grid maps discretise the environment into independent cells, neglecting important spatial correlations, and are unable to capture the continuous nature of the real world. With these drawbacks of grid maps in mind, Hilbert Maps (HM) and more recently Bayesian Hilbert Maps (BHMs), were introduced as a continuous representation of the environment. In this paper we propose a method to merge Bayesian Hilbert Maps built by a team of robots in a decentralised manner. The training of BHMs requires the inversion of a large covariance matrix, incurring cubic complexity. We introduce an approximation, Fast Bayesian Hilbert Maps (Fast-BHM), which reduces the time complexity to below quadratic. Such speed-ups allow the building and merging of Bayesian Hilbert Map models to be practical, opening the door for multi-robot Hilbert Map systems which can be much faster and more robust than an individual robot. By merging several individual Fast-BHMs in a decentralised manner we obtain a unified model of the environment which is itself a Fast-BHM. We conduct experiments to show that global Fast-BHM models do not deteriorate after repeated merging and training. We then empirically demonstrate, due to its the compact representation, fused Fast-BHMs outperform fusion methods involving discretising continuous representations, when the amount of information communicated is limited.

Deep Reinforcement Learning (DRL) has become a powerful methodology to solve complex decision-making problems. However, DRL has several limitations when used in real-world problems (e.g., robotics applications). For instance, long training times are required and cannot be accelerated in contrast to simulated environments, and reward functions may be hard to specify/model and/or to compute. Moreover, the transfer of policies learned in a simulator to the real-world has limitations (reality gap). On the other hand, machine learning methods that rely on the transfer of human knowledge to an agent have shown to be time efficient for obtaining well performing policies and do not require a reward function. In this context, we analyze the use of human corrective feedback during task execution to learn policies with high-dimensional state spaces, by using the D-COACH framework, and we propose new variants of this framework. D-COACH is a Deep Learning based extension of COACH (COrrective Advice Communicated by Humans), where humans are able to shape policies through corrective advice. The enhanced version of DCOACH, which is proposed in this paper, largely reduces the time and effort of a human for training a policy. Experimental results validate the efficiency of the D-COACH framework in three different problems (simulated and with real robots), and show that its enhanced version reduces the human training effort considerably, and makes it feasible to learn policies within periods of time in which a DRL agent do not reach any improvement.

Currently, automatically generated trajectories cannot be directly used on tasks that require high execution accuracies due to errors accused by inaccuracies in the robot model, actuator errors, and controller limitations. These trajectories often need manual refinement. This is not economically viable on low production volume applications. Unfortunately, execution errors are dependent on the nature of the trajectory and end-effector loads, and therefore devising a general purpose automated compensation scheme for reducing trajectory errors is not possible. This paper presents a method for analyzing the given trajectory, executing an exploratory physical run for a small portion of the given trajectory, and learning a compensation scheme based on the measured data. The learned compensation scheme is context-dependent and can be used to reduce the execution error. We have demonstrated the feasibility of this approach by conducting physical experiments.

In this paper the problem of contact-based navigation path planning for aerial robots is considered with the goal of enabling the autonomous in-contact operation on surfaces that can be highly anomalous. Such a capacity can prove critical in inspection through contact missions, as well as when a flying robot is tasked to operate in very narrow environments rendering safe free-flight impossible. To achieve this objective, beyond sliding in contact, a new locomotion primitive is introduced, namely that of azimuth rotations perpendicular to the surface under consideration. This new navigation mode, called flying cartwheel mode, offers navigation resourcefulness and resilience when the system is tasked to move in contact with surfaces that are otherwise non-traversable. The designed path planning method exploits both navigation modalities and a traversability metric to decide when to switch from sliding to flying cartwheel mode, and overall provides cost-optimal trajectories for in-contact navigation. The proposed approach is verified both in simulation, as well as experimentally using a surface presenting complex anomalies. It is highlighted that the proposed method does not assume any specialized contact mechanism or a control law tailored to physical interaction tasks, and hence is applicable to almost any micro aerial vehicle integrating protective shrouds around its propellers.

In this paper, we propose a contact-implicit trajectory optimization (CITO) method based on a variable smooth contact model (VSCM) and successive convexification (SCvx). The VSCM facilitates the convergence of gradient-based optimization without compromising physical fidelity. On the other hand, the proposed SCvx-based approach combines the advantages of direct and shooting methods for CITO. For evaluations, we consider non-prehensile manipulation tasks. The proposed method is compared to a version based on iterative linear quadratic regulator (iLQR) on a planar example. The results demonstrate that both methods can find physically-consistent motions that complete the tasks without a meaningful initial guess owing to the VSCM. The proposed SCvx-based method outperforms the iLQR-based method in terms of convergence, computation time, and the quality of motions found. Finally, the proposed SCvx-based method is tested on a standard robot platform and shown to perform efficiently for a real-world application.

This paper presents a contact-event-triggered filter using only a force-torque sensor with impedance control for non-cooperative, rotating, heavy object capture. Contact events are modeled for prediction, and detected to trigger the particle filter's updating process. By combining these features, a computationally efficient, contact-event-triggered filter is proposed. For our purpose of capture using impedance control, expected contact events, collisions and sliding are defined for prediction and detection. This novel method is implemented in an air bearing robotic system, and has demonstrated its superiority with the highest success rate (100%) for sliding contact mode cases, whereas the previous method could only yield a success rate of 87.9%. The computation resource is demonstrated to be limited, with a computation time of 4.2 milliseconds on average and 8.3 milliseconds at worst.

When performing tasks in uncertain environments and around humans, robots are likely to collide unexpectedly with people or objects. In order to ensure safety, most approaches rely on collision avoidance and try to prevent any contact from happening, which may result in unnecessary interruption of a task that would be feasible in spite of the obstacle. On the one hand, when an unexpected contact occurs, a safe robot behavior is required. On the other hand, it might be interesting to exploit the contact instead of moving away from it. In this paper, we present a contact-driven approach for safe and interactive robot operation to react to unforeseen contact events. This approach offers the possibility to control the contact while minimizing its effects on the robot tasks. It relies exclusively on the robot model and proprioceptive sensors. It is tested in simulation and hardware experiments on a 7 degrees of freedom robot arm and shows a safe contact behavior that does not interfere with the task, and as little as possible with the robot posture requirements.

Prioritized multi-objective optimization has been widely used within the operational space inverse dynamics control framework. In this paper, we present a constrained prioritized multi-objective optimization-base control formulation that extends to impedance control, including the `simple' impedance controller, which does not require the dynamic model. The main contribution of this paper is the dynamic-model-free prioritized feedback control formulation which encompasses arbitrary number of priority levels and takes the saturation constraints on the control inputs rigorously into account. The utility of the proposed formulation is demonstrated by a combined inverse dynamics impedance controller used to simulate stable locomotion of a planar anthropometric biped robot.