ICRA 2017 Papers

Robust environment perception is essential for decision-making on robots operating in complex domains. Intelligent task execution requires principled treatment of uncertainty sources in a robot's observation model. This is important not only for low-level observations (e.g., accelerom-eter data), but also for high-level observations such as semantic object labels. This paper formalizes the concept of macro-observations in Decentralized Partially Observable Semi-Markov Decision Processes (Dec-POSMDPs), allowing scalable semantic-level multi-robot decision making. A hierarchical Bayesian approach is used to model noise statistics of low-level classifier outputs, while simultaneously allowing sharing of domain noise characteristics between classes. Classification accuracy of the proposed macro-observation scheme, called Hierarchical Bayesian Noise Inference (HBNI), is shown to exceed existing methods. The macro-observation scheme is then integrated into a Dec-POSMDP planner, with hardware experiments running onboard a team of dynamic quadrotors in a challenging domain where noise-agnostic filtering fails. To the best of our knowledge, this is the first demonstration of a real-time, convolutional neural net-based classification framework running fully onboard a team of quadrotors in a multi-robot decision-making domain.

Autonomous robots that are to assist humans in their daily lives must recognize and understand the meaning of objects in their environment. However, the open nature of the world means robots must be able to learn and extend their knowledge about previously unknown objects on-line. In this work we investigate the problem of unknown object hypotheses generation, and employ a semantic Web-mining framework along with deep-learning-based object detectors. This allows us to make use of both visual and semantic features in combined hypotheses generation. Experiments on data from mobile robots in real world application deployments show that this combination improves performance over the use of either method in isolation.

Semantic mapping for autonomous mobile robots includes the place classification task that associates semantic labels (like `corridor' or `office') to rooms perceived in indoor environments. The mainstream approaches to place classification are characterized by local reasoning, where only features relative to the neighbourhood of each room are considered. In this paper, we propose a method for global reasoning on the whole structure of buildings, considered as single structured objects. We use a statistical relational learning algorithm, called kLog, and we compare it against a classifier, Extra-Trees, which resembles classical local approaches, in three tasks: classification of rooms, classification of entire floors of buildings, and validation of simulated worlds. Our results show that our global approach performs better than local approaches when the classification task involves reasoning on the regularities of buildings and when available information about rooms is coarse-grained.

Recognition of human manipulation actions together with the analysis and execution by a robot is an important issue. Also, perception of spatial relationships between objects is central to understanding the meaning of manipulation actions. Here we would like to merge these two notions and analyze manipulation actions using symbolic spatial relations between objects in the scene. Specifically, we define procedures for extraction of symbolic human-readable relations based on Axis Aligned Bounding Box object models and use sequences of those relations for action recognition from image sequences. Our framework is inspired by the so called Semantic Event Chain framework, which analyzes touching and un-touching events of different objects during the manipulation. However, our framework uses fourteen spatial relations instead of two. We show that our relational framework is able to differentiate between more manipulation actions than the original Semantic Event Chains. We quantitatively evaluate the method on the MANIAC dataset containing 120 videos of eight different manipulation actions and obtain 97% classification accuracy which is 12 % more as compared to the original Semantic Event Chains.

Future humanoid robots will be expected to carry out a wide range of tasks for which they had not been originally equipped by learning new skills and adapting to their environment. A crucial requirement towards that goal is to be able to take advantage of external elements as tools to perform tasks for which their own manipulators are insufficient; the ability to autonomously learn how to use tools will render robots far more versatile and simpler to design. Motivated by this prospect, this paper proposes and evaluates an approach to allow robots to learn tool affordances based on their 3D geometry. To this end, we apply tool-pose descriptors to represent tools combined with the way in which they are grasped, and affordance vectors to represent the effect tool-poses achieve in function of the action performed. This way, tool affordance learning consists in determining the mapping between these 2 representations, which is achieved in 2 steps. First, the dimensionality of both representations is reduced by unsupervisedly mapping them onto respective Self-Organizing Maps (SOMs). Then, the mapping between the neurons in the tool-pose SOM and the neurons in the affordance SOM for pairs of tool-poses and their corresponding affordance vectors, respectively, is learned with a neural based regression model. This method enables the robot to accurately predict the effect of its actions using tools, and thus to select the best action for a given goal, even with tools not seen on the learning phase.

Accurate road segmentation is a prerequisite for autonomous driving. Current state-of-the-art methods are mostly based on convolutional neural networks (CNNs). Nevertheless, their good performance is at expense of abundant annotated data and high computational cost. In this work, we address these two issues by a self-paced cross-modality transfer learning framework with efficient projection CNN. To be specific, with the help of stereo images, we first tackle a relevant but easier task, i.e. free-space detection with well developed unsupervised methods. Then, we transfer these useful but noisy knowledge in depth modality to single RGB modality with self-paced CNN learning. Finally, we only need to fine-tune the CNN with a few annotated images to get good performance. In addition, we propose an efficient projection CNN, which can improve the fine-grained segmentation results with little additional cost. At last, we test our method on KITTI road benchmark. Our proposed method surpasses all published methods at a speed of 15fps.

This paper describes additive self-folding, an origami-inspired rapid fabrication approach for creating actuatable compliant structures. Recent work in 3-D printing and other rapid fabrication processes have mostly focused on rigid objects or objects that can achieve small deformations. In contrast, soft robots often require elastic materials and large amounts of movement. Additive self-folding is a process that involves cutting slices of a 3-D object in a long strip and then pleat folding them into a likeness of the original model. The zigzag pattern for folding enables large bending movements that can be actuated and controlled. Gaps between slices in the folded model can be designed to provide larger deformations or higher shape accuracy. We advance existing planar fabrication and self-folding techniques to automate the fabrication process, enabling highly compliant structures with complex 3-D geometries to be designed and fabricated within a few hours. We describe this process in this paper and provide algorithms for converting 3-D meshes into additive self-folding designs. The designs can be rapidly instrumented for global control using magnetic fields or tendon-driven for local bending. We also describe how the resulting structures can be modeled and their responses to tendon-driven control predicted. We test our design and fabrication methods on three models (a bunny, a tuna fish, and a starfish) and demonstrate the method's potential for actuation by actuating the tuna fish and starfish models using tendons and magnetic control.

Place recognition in 3D data is a challenging task that has been commonly approached by adapting image-based solutions. Methods based on local features suffer from ambiguity and from robustness to environment changes while methods based on global features are viewpoint dependent. We propose SegMatch, a reliable place recognition algorithm based on the matching of 3D segments. Segments provide a good compromise between local and global descriptions, incorporating their strengths while reducing their individual drawbacks. SegMatch does not rely on assumptions of `perfect segmentation', or on the existence of `objects' in the environment, which allows for reliable execution on large scale, unstructured environments. We quantitatively demonstrate that SegMatch can achieve accurate localization at a frequency of 1Hz on the largest sequence of the KITTI odometry dataset. We furthermore show how this algorithm can reliably detect and close loops in real-time, during online operation. In addition, the source code for the SegMatch algorithm is made publicly available.

Fluidic actuators allow versatile, agile, and powerful motions and are commonly applied in robotics and automation. Likewise, many biological systems use fluidic actuators implemented with tissue for a wealth of tasks and performances. Spiders for example apply a hybrid mechanism of hydraulically actuated joint extension and muscle-based joint flexion to produce movement in two of their seven leg joints. Here, we present a novel spider-inspired joint mechanism employing both pneumatics and electrically-actuated tendons capable of strong, dynamic, and rapid joint movement. The implementation of the joint is closely inspired by those seen in real spiders, with a foldable structured membrane that effectively transfers all the energy from pressure to torque as the leg unfolds. To evaluate the mechanism we derived static joint models and a simple jumping model, and conducted equivalent experimental tests with a prototype of a single jumping leg robot. Besides applications in robot locomotion, the implementation and modeling of the spider-inspired joint mechanism can be utilized to further explore dynamics and functional biomechanics in spiders. In the future, we hope to use this platform to answer questions related to the impressive jumping and locomotion performances of real arachnids, and explore what morphological traits lie behind efficient spider locomotion at different size scales.

This paper presents the first ever approach for solving continuous-observation Decentralized Partially Observable Markov Decision Processes (Dec-POMDPs) and their semi-Markovian counterparts, Dec-POSMDPs. This contribution is especially important in robotics, where a vast number of sensors provide continuous observation data. A continuous-observation policy representation is introduced using Stochastic Kernel-based Finite State Automata (SK-FSAs). An SK-FSA search algorithm titled Entropy-based Policy Search using Continuous Kernel Observations (EPSCKO) is introduced and applied to the first ever continuous-observation Dec-POMDP/Dec-POSMDP domain, where it significantly outperforms state-of-the-art discrete approaches. This methodology is equally applicable to Dec-POMDPs and Dec-POSMDPs, though the empirical analysis presented focuses on Dec-POSMDPs due to their higher scalability. To improve convergence, an entropy injection policy search acceleration approach for both continuous and discrete observation cases is also developed and shown to improve convergence rates without degrading policy quality.

We propose a vision-based method that localizes a ground vehicle using publicly available satellite imagery as the only prior knowledge of the environment. Our approach takes as input a sequence of ground-level images acquired by the vehicle as it navigates, and outputs an estimate of the vehicle's pose relative to a georeferenced satellite image. We overcome the significant viewpoint and appearance variations between the images through a neural multi-view model that learns location-discriminative embeddings in which ground-level images are matched with their corresponding satellite view of the scene. We use this learned function as an observation model in a filtering framework to maintain a distribution over the vehicle's pose. We evaluate our method on different benchmark datasets and demonstrate its ability localize ground-level images in environments novel relative to training, despite the challenges of significant viewpoint and appearance variations.

This paper reports on an algorithm for planning trajectories that allow a multirotor micro aerial vehicle (MAV) to quickly identify a set of unknown parameters. In many problems like self calibration or model parameter identification some states are only observable under a specific motion. These motions are often hard to find, especially for inexperienced users. Therefore, we consider system model identification in an active setting, where the vehicle autonomously decides what actions to take in order to quickly identify the model. Our algorithm approximates the belief dynamics of the system around a candidate trajectory using an extended Kalman filter (EKF). It uses sampling-based motion planning to explore the space of possible beliefs and find a maximally informative trajectory within a user-defined budget. We validate our method in simulation and on a real system showing the feasibility and repeatability of the proposed approach. Our planner creates trajectories which reduce model parameter convergence time and uncertainty by a factor of four.

In this paper, we propose a sampling-based policy iteration for optimal planning under temporal logic constraints. The method integrates approximate optimal control, importance sampling, and formal methods. For a subclass of linear temporal logic, the planning problem is transformed to an optimal control problem for a hybrid system where discrete transitions are triggered by linear time events in temporal logic. Instead of solving the Hamilton-Jacobi-Bellman equation, we use policy function approximation to reduce the problem into a search of an optimal weight vector that parametrizes the near-optimal policy for given bases. Then, we incorporate Model Reference Adaptive Search - an importance sampling-based optimization algorithm to perform a sample-efficient search within the parameter space of policy function approximations. Facing the discontinuity in cost function introduced by temporal logic constraints and system dynamics, we introduce (1) a rank function in formal logic specifications to enable sample-efficient search; (2) specification-guided basis selection. Under mild technical assumptions, the proposed algorithm converges, with probability one, to a global approximate optimal policy that ensures the satisfaction of temporal logic constraints. The correctness and efficiency of the method are demonstrated through numerical experiments including temporal logic planning for a linear system and a nonlinear mobile robot.

Motion planning under differential constraints is one of the canonical problems in robotics. State-of-the-art methods evolve around kinodynamic variants of popular sampling-based algorithms, such as Rapidly-exploring Random Trees (RRTs). However, there are still challenges remaining, for example, how to include complex dynamics while guaranteeing optimality. If the open-loop dynamics are unstable, exploration by random sampling in control space becomes inefficient. We describe CL-RRT#, which leverages ideas from the RRT# algorithm and a variant of the RRT algorithm, which generates trajectories using closed-loop prediction. Planning with closed-loop prediction allows us to handle complex unstable dynamics and avoids the need to find computationally hard steering procedures. The search technique presented in the RRT# algorithm allows us to improve the solution quality by searching over alternative reference trajectories. We show the benefits of the proposed approach on an autonomous-driving scenario.

In multi-robot systems where a central decision maker is specifying the movement of each individual robot, a communication failure can severely impair the performance of the system. This paper develops a motion strategy that allows robots to safely handle critical communication failures for such multi-robot architectures. For each robot, the proposed algorithm computes a time horizon over which collisions with other robots are guaranteed not to occur. These safe time horizons are included in the commands being transmitted to the individual robots. In the event of a communication failure, the robots execute the last received velocity commands for the corresponding safe time horizons leading to a provably safe open-loop motion strategy. The resulting algorithm is computationally effective and is agnostic to the task that the robots are performing. The efficacy of the strategy is verified in simulation as well as on a team of differential-drive mobile robots.

This paper presents TIREBOT, a novel tire-workshop robotic co-worker that can safely move in a tire workshop and assist the operator in lifting and transporting wheels among several working stations. A safe and cooperative navigation strategy based on the concept of danger field is illustrated. Finally, TIREBOT is experimentally evaluated in a real tire-workshop and used by real operator in order to assess the usability and the effectiveness of the robot in a real operating scenario.

In this paper, we consider a networked multi-robot system operating in an obstacle populated planar workspace under a single leader-multiple followers architecture. We propose a decentralized reconfiguration strategy of the set of connectivity and formation specifications that assures convergence to the desired point, while guaranteeing global connectivity. In particular, we construct a low-level Decentralized Navigation Functions based controller that encodes the goals and safety requirements of the system. However, owing to topological obstructions, stable critical points other than the desired one may appear. In such case, we employ a high-level distributed discrete procedure which attempts to solve a Distributed Constraint Satisfaction Problem on a local Voronoi partition, providing the necessary reconfiguration for the system to progress towards its goal. Eventually, we show that the system either converges to the desired point or attains a tree configuration with respect to the formation topology, in which case the system switches to a novel controller based on the Prescribed Performance technique, that eventually guarantees convergence. Finally, a simulation study clarifies and verifies the approach.

Safety Barrier Certificates that ensure collision-free maneuvers for teams of differential flatness-based quadrotors are presented in this paper. Synthesized with control barrier functions, the certificates are used to modify the nominal trajectory in a minimally invasive way to avoid collisions. The proposed collision avoidance strategy complements existing flight control and planning algorithms by providing trajectory modifications with provable safety guarantees. The effectiveness of this strategy is supported both by the theoretical results and experimental validation on a team of five quadrotors.

Robotic systems typically require memory recall mechanisms for a variety of tasks including localization, mapping, planning, visualization etc. We argue for a novel memory recall framework that enables more complex inference schemas by separating the computation from its associated data. In this work we propose a shared, centralized data persistence layer that maintains an ensemble of online, situationally-aware robot states. This is realized through a queryable graph-database with an accompanying key-value store for larger data. In turn, this approach is scalable and enables a multitude of capabilities such as experience-based learning and long-term autonomy. Using multi-modal simultaneous localization and mapping and a few example use-cases, we demonstrate the versatility and extensible nature that centralized persistence and SLAMinDB can provide. In order to support the notion of life-long autonomy, we envision robots to be endowed with such a persistence model, enabling them to revisit previous experiences and improve upon their existing task-specific capabilities.