RSS 2020 Papers

We present Nav2Goal, a data-efficient and end-to-end learning method for goal-conditioned visual navigation. Our technique is used to train a navigation policy that enables a robot to navigate close to sparse geographic waypoints provided by a user without any prior map, all while avoiding obstacles and choosing paths that cover user-informed regions of interest. Our approach is based on recent advances in conditional imitation learning. General-purpose safe and informative actions are demonstrated by a human expert. The learned policy is subsequently extended to be goal-conditioned by training with hindsight relabelling, guided by the robot's relative localization system, which requires no additional manual annotation. We deployed our method on an underwater vehicle in the open ocean to collect scientifically relevant data of coral reefs, which allowed our robot to operate safely and autonomously, even at very close proximity to the coral. Our field deployments have demonstrated over a kilometer of autonomous visual navigation, where the robot reaches on the order of 40 waypoints, while collecting scientifically relevant data. This is done while travelling within 0.5 m altitude from sensitive corals and exhibiting significant learned agility to overcome turbulent ocean conditions and to actively avoid collisions.

In this work, we present a spiking neural network (SNN) based PID controller on a neuromorphic chip. On-chip SNNs are currently being explored in low-power AI applications. Due to potentially ultra-low power consumption, low latency, and high processing speed, on-chip SNNs are a promising tool for control of power-constrained platforms, such as Unmanned Aerial Vehicles (UAV). To obtain highly efficient and fast end-to-end neuromorphic controllers, the SNN-based AI architectures must be seamlessly integrated with motor control. Towards this goal, we present here the first implementation of a fully neuromorphic PID controller. We interfaced Intel's neuromorphic research chip Loihi to a UAV, constrained to a single degree of freedom. We developed an SNN control architecture using populations of spiking neurons, in which each spike carries information about the measured, control, or error value, defined by the identity of the spiking neuron. Using this sparse code, we realize a precise PID controller. The P, I, and D gains of the controller are implemented as synaptic weights that can adapt according to an on-chip plasticity rule. In future work, these plastic synapses can be used to tune and adapt the controller autonomously.

Embodiment is an important characteristic for all intelligent agents (creatures and robots), while existing scene description tasks mainly focus on analyzing images passively and the semantic understanding of the scenario is separated from the interaction between the agent and the environment. In this work, we propose the \textit{Embodied Scene Description}, which exploits the embodiment ability of the agent to find an optimal viewpoint in its environment for scene description tasks. A learning framework with the paradigms of imitation learning and reinforcement learning is established to teach the intelligent agent to generate corresponding sensorimotor activities. The proposed framework is tested on both the AI2Thor dataset and a real world robotic platform demonstrating the effectiveness and extendability of the developed method.

We introduce a reconfigurable underactuated robot hand able to perform systematic prehensile in-hand manipulations regardless of object size or shape. The hand utilises a two-degree-of-freedom five-bar linkage as the palm of the gripper, with three three-phalanx underactuated fingers---jointly controlled by a single actuator---connected to the mobile revolute joints of the palm. Three actuators are used in the robot hand system, one for controlling the force exerted on objects by the fingers and two for changing the configuration of the palm. This novel layout allows decoupling grasping and manipulation, facilitating the planning and execution of in-hand manipulation operations. The reconfigurable palm provides the hand with large grasping versatility, and allows easy computation of a map between task space and joint space for manipulation based on distance-based linkage kinematics. The motion of objects of different sizes and shapes from one pose to another is then straightforward and systematic, provided the objects are kept grasped. This is guaranteed independently and passively by the underactuated fingers using a custom tendon routing method, which allows no tendon length variation when the relative finger base position changes with palm reconfigurations. We analyse the theoretical grasping workspace and manipulation capability of the hand, present algorithms for computing the manipulation map and in-hand manipulation planning, and evaluate all these experimentally. Numerical and empirical results of several manipulation trajectories with objects of different size and shape clearly demonstrate the viability of the proposed concept.

In the past years, research on the embodiment of interactive social agents has been focused on comparisons between robots and virtually-displayed agents. Our work contributes to this line of research by providing a comparison between social robots and disembodied agents exploring the role of embodiment within group interactions. We conducted a user study where participants formed a team with two agents to play a Collective Risk Dilemma (CRD). Besides having two levels of embodiment as between-subjects ---physically-embodied and disembodied---, we also manipulated the agents' degree of cooperation as a within-subjects variable ---one of the agents used a prosocial strategy and the other used selfish strategy. Our results show that while trust levels were similar between the two conditions of embodiment, participants identified more with the team of embodied agents. Surprisingly, when the agents were disembodied, the prosocial agent was rated more positively and the selfish agent was rated more negatively, compared to when they were embodied. The obtained results support that embodied interactions might improve how humans relate with agents in team settings. However, if the social aspects can positively mask selfish behaviours, as our results suggest, a dark side of embodiment may emerge.

Truly intelligent agents need to capture the interplay of all their senses to build a rich physical understanding of their world. In robotics, we have seen tremendous progress in using visual and tactile perception; however we have often ignored a key sense: sound. This is primarily due to lack of data that captures the interplay of action and sound. In this work, we perform the first large-scale study of the interactions between sound and robotic action. To do this, we create the largest available sound-action-vision dataset with 15,000 interactions on60 objects using our robotic platform Tilt-Bot. By tilting objects and allowing them to crash into the walls of a robotic tray, we collect rich four-channel audio information. Using this data, we explore the synergies between sound and action, and present three key insights. First, sound is indicative of fine-grained object class information, e.g., sound can differentiate a metal screwdriver from a metal wrench. Second, sound also contains information about the causal effects of an action, i.e. given the sound produced, we can predict what action was applied on the object. Finally, object representations derived from audio embeddings are indicative of implicit physical properties. We demonstrate that on previously unseen objects, audio embeddings generated through interactions can predict forward models 24%better than passive visual embeddings.

There is a rising interest in Spatio-temporal systems described by Partial Differential Equations (PDEs) among the control community. Not only are these systems challenging to control, but the sizing and placement of their actuation is an NP-hard problem on its own. Recent methods either discretize the space before optimziation, or apply tools from linear systems theory under restrictive linearity assumptions. In this work we consider control and actuator placement as a coupled optimization problem, and derive an optimization algorithm on Hilbert spaces for nonlinear PDEs with an additive spatio-temporal description of white noise. We study first and second order systems and in doing so, extend several results to the case of second order PDEs. The described approach is based on variational optimization, and performs joint RL-type optimization of the feedback control law and the actuator design over episodes. We demonstrate the efficacy of the proposed approach with several simulated experiments on a variety of SPDEs.

Typical end-to-end formulations for learning robotic navigation involve predicting a small set of steering command actions (e.g., step forward, turn left, turn right, etc.) from images of the current state (e.g., a bird's-eye view of a SLAM reconstruction). Instead, we show that it can be advantageous to learn with dense action representations defined in the same domain as the state. In this work, we present "spatial action maps," in which the set of possible actions is represented by a pixel map (aligned with the input image of the current state), where each pixel represents a local navigational endpoint at the corresponding scene location. Using ConvNets to infer spatial action maps from state images, action predictions are thereby spatially anchored on local visual features in the scene, enabling significantly faster learning of complex behaviors for mobile manipulation tasks with reinforcement learning. In our experiments, we task a robot with pushing objects to a goal location, and find that policies learned with spatial action maps achieve much better performance than traditional alternatives.

We present a numerical method to compute singularity sets in the configuration space of free-floating robots, comparing two different criteria based on formal methods. By exploiting specific properties of free-floating systems and an alternative formulation of the generalized Jacobian, the search space and computational complexity of the algorithm is reduced. It is shown that the resulting singularity maps can be applied in the context of trajectory planning to guarantee feasibility with respect to singularity avoidance. The proposed approach is validated on a space robot composed of a six degrees-of-freedom (DOF) arm mounted on a body with six DOF.

Enabling robots to learn tasks and follow instructions as easily as humans is important for many real-world robot applications. Previous approaches have applied machine learning to teach the mapping from language to low dimensional symbolic representations constructed by hand, using demonstration trajectories paired with accompanying instructions. These symbolic methods lead to data efficient learning. Other methods map language directly to high-dimensional control behavior, which requires less design effort but is data-intensive. We propose to first learning symbolic abstractions from demonstration data and then mapping language to those learned abstractions. These symbolic abstractions can be learned with significantly less data than end-to-end approaches, and support partial behavior specification via natural language since they permit planning using traditional planners. During training, our approach requires only a small number of demonstration trajectories paired with natural language—without the use of a simulator—and results in a representation capable of planning to fulfill natural language instructions specifying a goal or partial plan. We apply our approach to two domains, including a mobile manipulator, where a small number of demonstrations enable the robot to follow navigation commands like “Take left at the end of the hallway,” in environments it has not encountered before.

In this paper, we introduce and tackle the simultaneous enhancement and super-resolution (SESR) problem for underwater robot vision and provide an efficient solution for near real-time applications. We present Deep SESR, a residual-in-residual network-based generative model that can learn to restore perceptual image qualities at 2x, 3x, or 4x higher spatial resolution. We supervise its training by formulating a multi-modal objective function that addresses the chrominance-specific underwater color degradation, lack of image sharpness, and loss in high-level feature representation. It is also supervised to learn salient foreground regions in the image, which in turn guides the network to learn global contrast enhancement. We design an end-to-end training pipeline to jointly learn the saliency prediction and SESR on a shared hierarchical feature space for fast inference. Moreover, we present UFO-120, the first dataset to facilitate large-scale SESR learning; it contains over 1500 training samples and a benchmark test set of 120 samples. By thorough experimental evaluation on UFO-120 and several other standard datasets, we demonstrate that Deep SESR outperforms the existing solutions for underwater image enhancement and super-resolution. We also validate its generalization performance on several test cases that include underwater images with diverse spectral and spatial degradation levels, and also terrestrial images with unseen natural objects. Lastly, we analyze its computational feasibility for single-board deployments and demonstrate its operational benefits for visually-guided underwater robots.

Assistive robots enable people with disabilities to conduct everyday tasks on their own. However, these tasks can be complex, containing both coarse reaching motions and fine-grained manipulation. For example, when eating, not only does one need to move to the correct food item, but they must also precisely manipulate the food in different ways (e.g., cutting, stabbing, scooping). Shared autonomy methods make robot teleoperation safer and more precise by arbitrating user inputs with robot controls. However, these works have focused mainly on the high-level task of reaching a goal from a discrete set, while largely ignoring manipulation of objects at that goal. Meanwhile, dimensionality reduction techniques for teleoperation map useful high-dimensional robot actions into an intuitive low-dimensional controller, but it is unclear if these methods can achieve the requisite precision for tasks like eating. Our insight is that---by combining intuitive embeddings from learned latent actions with robotic assistance from shared autonomy---we can enable precise assistive manipulation. In this work, we adopt learned latent actions for shared autonomy by proposing a new model structure that changes the meaning of the human's input based on the robot's confidence of the goal. We show convergence bounds on the robot's distance to the most likely goal, and develop a training procedure to learn a controller that is able to move between goals even in the presence of shared autonomy. We evaluate our method in simulations and an eating user study. See videos of our experiments here: https://youtu.be/7BouKojzVyk.

Publicly available satellite imagery can be an ubiquitous, cheap, and powerful tool for vehicle localisation when a prior sensor map is unavailable. However, satellite images are not directly comparable to data from ground range sensors because of their starkly different modalities. We present a learned metric localisation method that not only handles the modality difference, but is cheap to train, learning in a self-supervised fashion without metrically accurate ground truth. By evaluating across multiple real-world datasets, we demonstrate the robustness and versatility of our method for various sensor configurations. We pay particular attention to the use of millimetre wave radar, which, owing to its complex interaction with the scene and its immunity to weather and lighting, makes for a compelling and valuable use case.

Modular robotic systems comprise groups of physically connected modules which can be reconfigured to create morphologies that suit an environment or task. One method of reconfiguration is via subtraction, where extraneous modules disconnect from an initial configuration, before being removed by external intervention. In this paper, we consider an approach to reconfiguration in two dimensions, here termed active subtraction, in which unwanted modules traverse a configuration in order to remove themselves safely, without the need for external intervention, making it a form of self-reconfiguration. We present a sequential solution that selects suitable extraneous modules that then remove themselves, one by one. We also present a parallel solution that, while being more computationally demanding, allows multiple modules to move simultaneously. Both solutions are proven to (i) be correct for any given non-hollow structure, and (ii) require, in the worst case, quadratic time proportionally to the number of modules. Simulation studies demonstrate that both solutions work effectively for specified and randomly generated desired configurations with hundreds of modules, and reveal a non-monotonic dependence between the performance and the percentage of modules to be removed. This work demonstrates active subtraction as a viable method of self-reconfiguration, without the need for heuristics or stochasticity, and suggests its potential for application in real-world systems.

By harnessing a growing dataset of robot experience, we learn control policies for a diverse and increasing set of related manipulation tasks. To make this possible, we introduce reward sketching: an effective way of eliciting human preferences to learn the reward function for a new task. This reward function is then used to retrospectively annotate all historical data, collected for different tasks, with predicted rewards for the new task. The resulting massive annotated dataset can then be used to learn manipulation policies with batch reinforcement learning (RL) from visual input in a completely off-line way, i.e., without interactions with the real robot. This approach makes it possible to scale up RL in robotics, as we no longer need to run the robot for each step of learning. We show that the trained batch RL agents, when deployed in real robots, can perform a variety of challenging tasks involving multiple interactions among rigid or deformable objects. Moreover, they display a significant degree of robustness and generalization. In some cases, they even outperform human teleoperators.

There is increasing demand for automated systems that can fabricate 3D structures. Robotic spatial extrusion has become an attractive alternative to traditional layer-based 3D printing due to a manipulator's flexibility to print large, directionally-dependent structures. However, existing extrusion planning algorithms require a substantial amount of human input, do not scale to large instances, and lack theoretical guarantees. In this work, we present a rigorous formalization of robotic spatial extrusion planning and provide several efficient and probabilistically complete planning algorithms. The key planning challenge is, throughout the printing process, satisfying both stiffness constraints that limit the deformation of the structure and geometric constraints that ensure the robot does not collide with the structure. We show that, although these constraints often conflict with each other, a greedy backward state-space search guided by a stiffness-aware heuristic is able to successfully balance both constraints. We empirically compare our methods on a benchmark of over 40 simulated extrusion problems. Finally, we apply our approach to 3 real-world extrusion problems.

This paper presents a game-theoretic path-following formulation where the opponent is an adversary road model. This formulation allows us to compute safe sets using tools from viability theory, that can be used as terminal constraints in an optimization-based motion planner. Based on the adversary road model, we first derive an analytical discriminating domain, which even allows guaranteeing safety in the case when steering rate constraints are considered. Second, we compute the discriminating kernel and show that the output of the gridding based algorithm can be accurately approximated by a fully connected neural network, which can again be used as a terminal constraint. Finally, we show that by using our proposed safe sets, an optimization-based motion planner can successfully drive on city and country roads with prediction horizons too short for other baselines to complete the task.

The multiple-path orienteering problem asks for paths for a team of robots that maximize the total reward collected while satisfying budget constraints on the path length. This problem models many multi-robot routing tasks such as exploring unknown environments and information gathering for environmental monitoring. In this paper, we focus on how to make the robot team robust to failures when operating in adversarial environments. We introduce the Robust Multiple path Orienteering Problem (RMOP) where we seek worst-case guarantees against an adversary that is capable of attacking at most \alpha robots. Our main contribution is a general approximation scheme with bounded approximation guarantee that depends on \alpha and the approximation factor for single robot orienteering. In particular, we show that the algorithm yields a (i) constant factor approximation when the cost function is modular; (ii) log factor approximation when the cost function is submodular; and (iii) constant-factor approximation when the cost function is submodular but the robots are allowed to exceed their path budgets by a bounded amount. In addition to theoretical analysis, we perform simulation study for an ocean monitoring application to demonstrate the efficacy of our approach.

Natural language object retrieval is a highly useful yet challenging task for robots in human-centric environments. Previous work has primarily focused on commands specifying the desired object's type such as "scissors" and/or visual attributes such as "red," thus limiting the robot to only known object classes. We develop a model to retrieve objects based on descriptions of their usage. The model takes in a language command containing a verb, for example "Hand me something to cut," and RGB images of candidate objects and selects the object that best satisfies the task specified by the verb. Our model directly predicts an object's appearance from the object's use specified by a verb phrase. We do not need to explicitly specify an object's class label. Our approach allows us to predict high level concepts like an object's utility based on the language query. Based on contextual information present in the language commands, our model can generalize to unseen object classes and unknown nouns in the commands. Our model correctly selects objects out of sets of five candidates to fulfill natural language commands, and achieves an average accuracy of 62.3% on a held-out test set of unseen ImageNet object classes and 53.0% on unseen object classes and unknown nouns. Our model also achieves an average accuracy of 54.7% on unseen YCB object classes, which have a different image distribution from ImageNet objects. We demonstrate our model on a KUKA LBR iiwa robot arm, enabling the robot to retrieve objects based on natural language descriptions of their usage. We also present a new dataset of 655 verb-object pairs denoting object usage over 50 verbs and 216 object classes.