While humans can successfully navigate using abstractions, ignoring details that are irrelevant to the task at hand, most of the existing approaches in robotics require detailed environment representations which consume a significant amount of sensing, computing, and storage; these issues become particularly important in resource-constrained settings with limited power budgets. Deep learning methods can learn from prior experience to abstract knowledge from novel environments, and use it to more efficiently execute tasks such as frontier exploration, object search, or scene understanding. We propose BoxMap, a Detection-Transformer-based architecture that takes advantage of the structure of the sensed partial environment to update a topological graph of the environment as a set of semantic entities (rooms and doors) and their relations (connectivity). The predictions from low-level measurements can be leveraged to achieve high-level goals with lower computational costs than methods based on detailed representations. As an example application, we consider a robot equipped with a 2-D laser scanner tasked with exploring a residential building. Our BoxMap representation scales quadratically with the number of rooms (with a small constant), resulting in significant savings over a full geometric map. Moreover, our high-level topological representation results in 30.9 % shorter trajectories in the exploration task with respect to a standard method. Code is available at: bit.ly/3F6w2Yl.

Deep learning methods are widely used in robotic applications. By learning from prior experience, the robot can abstract knowledge of the environment, and use this knowledge to accomplish different goals, such as object search, frontier exploration, or scene understanding, with a smaller amount of resources than might be needed without that knowledge. Most existing methods typically require a significant amount of sensing, which in turn has significant costs in terms of power consumption for acquisition and processing, and typically focus on models that are tuned for each specific goal, leading to the need to train, store and run each one separately. These issues are particularly important in a resource-constrained setting, such as with small-scale robots or during long-duration missions. We propose a single, multi-task deep learning architecture that takes advantage of the structure of the partial environment to predict different abstractions of the environment (thus reducing the need for rich sensing), and to leverage these predictions to simultaneously achieve different high-level goals (thus sharing computation between goals). As an example application of the proposed architecture, we consider the specific example of a robot equipped with a 2-D laser scanner and an object detector, tasked with searching for an object (such as an exit) in a residential building while constructing a topological map that can be used for future missions. The prior knowledge of the environment is encoded using a U-Net deep network architecture. In this context, our work leads to an object search algorithm that is complete, and that outperforms a more traditional frontier-based approach. The topological map we produce uses scene trees to qualitatively represent the environment as a graph at a fraction of the cost of existing SLAM-based solutions. Our results demonstrate that it is possible to extract multi-task semantic information that is useful for navigation and mapping directly from bare-bone, non-semantic measurements.

This work focuses on the development of an algorithm to design sampling patterns for acquiring images in atomic force microscopy (AFM). The goal is to reduce data acquisition time of the instrument without significantly compromising image quality by appropriately allocating sampling effort. We assume the frequency structure of the images to be sampled, and in particular the locations of the large frequency coefficients, is partially known. Based on this knowledge, the expected reconstruction error using simplified matching pursuit (SMP) is analyzed and a Monte Carlo-based strategy is proposed to create sampling patterns that minimize the expected error. The proposed method is demonstrated through simulation on two groups of sample surface images from two different materials acquired by AFM.

In this paper, we consider a robot motion planning problem from a specification given as a syntactically co-safe linear temporal logic formula over a set of properties known to be satisfied at the regions of an unknown environment. The robot is assumed to be equipped with deterministic motion and accurate sensing capabilities. The environment is assumed to be partitioned into a finite number of identical square cells. By bringing together tools from formal verification, graph theory, and grid-based exploration, we develop an incremental algorithm that makes progress towards satisfying the specification while the robot discovers the environment using its local sensors. We show that the algorithm is sound and complete. We illustrate the feasibility and effectiveness of our approach through a simulated case study.

The increasing need for real time robotic systems capable of performing tasks in changing and constrained environments demands the development of reliable and adaptable motion planning and control algorithms. This paper considers a mobile robot whose performance is measured by the completion of temporal logic tasks within a certain period of time. In addition to such time constraints, the planning algorithm must also deal with changes in the robot's workspace during task execution. In our case, the robot is deployed in a partitioned environment subjected to structural changes in which doors shift from open to closed and vice-versa. The motion of the robot is modeled as a Continuous Time Markov Decision Process and the robot's mission is expressed as a Continuous Stochastic Logic (CSL) temporal logic specification. An approximate solution to find a control strategy that satisfies such specifications is derived for a subset of probabilistic CSL formulae. Simulation and experimental results are provided to illustrate the method.

We describe a computational framework for automatic deployment of a robot with sensor and actuator noise from a temporal logic specification over a set of properties that are satisfied by the regions of a partitioned environment. We model the motion of the robot in the environment as a Markov decision process (MDP) and translate the motion specification to a formula of probabilistic computation tree logic (PCTL). As a result, the robot control problem is mapped to that of generating an MDP control policy from a PCTL formula. We present algorithms for the synthesis of such policies for different classes of PCTL formulas. We illustrate our method with simulation and experimental results.

Mobile robotic systems move in environments that are constantly changing due to the presence of dynamic obstacles. In this work we consider one such environment in which the dynamic nature comes from doors that can open or close during the robot's mission. We derive a solution to the automatic deployment of a robot from a temporal logic specification assuming three different levels of knowledge and sensing capabilities of the robot. Under each one of these settings, the motion of the robot is modeled either as a Markov decision process (MDP) or mixed observability MDP. The objective is to find a control strategy that maximizes the probability of satisfying a specification given in Probabilistic Computation Tree Logic (PCTL). We describe an optimal solution in one setting and sub-optimal, reactive solutions in the other two. We illustrate our methods with simulation results.

We present a computational framework for automatic deployment of a robot from a temporal logic specification over a set of properties of interest satisfied at the regions of a partitioned environment. We assume that, during the motion of the robot in the environment, the current region can be precisely determined, while due to sensor and actuation noise, the outcome of a control action can only be predicted probabilistically. Under these assumptions, the deployment problem translates to generating a control strategy for a Markov Decision Process (MDP) from a temporal logic formula.We propose an algorithm inspired from probabilistic Computation Tree Logic (PCTL) model checking to find a control strategy that maximizes the probability of satisfying the specification. We illustrate our method with simulation and experimental results.

We present a computational framework and experimental setup for deployment of autonomous cars in a miniature Robotic Urban-Like Environment (RULE). The specifications are given in rich, human-like language as temporal logic statements about roads, intersections, and parking spaces. We use transition systems to model the motion and sensing capabilities of the robots and the topology of the environment and use tools resembling model checking to generate robot control strategies and to verify the correctness of the solution. The experimental setup is based on Khepera III robots, which move autonomously on streets while observing traffic rules.

We consider the problem of approximating a finite-length continuous curve by a piecewise linear one whose segments are assumed to be connected by 2 DOF joints. We solve the problem under the assumption that the endpoints of the line segments lie on the continuous curve. Analytical expressions for the relative orientations of each pair of line segments as a function of a single rotational DOF are found. This angle can be chosen arbitrarily or used to optimize a secondary task. The motivating application for this paper is the control of a snake-like robot using gaits designed from shape primitives.

We consider the problem of approximating a continuous curve by a piecewise linear one whose segments are assumed to be connected by universal joints. Rather than taking a least-squares approach, we require that the endpoints of the line segments lie on the continuous curve. We show that under these assumptions a single rotational degree of freedom remains. An algorithm is derived to determine the set of angles characterizing the relative orientation of each consecutive pair of line segments as a function of this rotational degree of freedom. Two examples are given to illustrate the procedure. The motivating application for this work is the control of a snake-like robot using a set of gaits designed from shape primitives

Landmark-based graphs are a useful and parsimonious tool for representing large scale environments. Relating landmarks by means of feedback-control algorithms encoded in a motion description language provides a level of abstraction that enables autonomous vehicles to navigate effectively by composing strings in the language to form complex strategies that would be difficult to design at the level of sensors and actuators. In such a setting, feedback control requires one to pay attention not only to sensor and actuator uncertainty, but also to the ambiguity introduced by the fact that many landmarks may look similar when using a modest set of observations. This work discusses the generation of language-based feedback control sequences for landmark-based navigation together with the problem of sensing landmarks sufficiently well to make feedback meaningful. The paper makes two contributions. First, we extend previous work to include the costs of sensing with varying degrees of accuracy. Second, we describe a Monte Carlo based approach to landmark sensing which relies on the use of particle filters. We include simulation results that illustrate our approach.

We address two key goals pertaining to autonomous mobile robots: one, to develop fast accurate sensory capabilities - at present, the localization of sound sources - and second, the integration of such sensory modules with other robot functions, especially its motor control and navigation. A primary motivation for this work was to devise effective means to guide robotic navigation in environments with acoustic sources. We designed and built a biomimetic sound-source localization apparatus. In contrast to the popular use of time-of-arrival differences in free field microphone arrays, our system is based on the principles observed in nature, where directional acoustic sensing evolved to rely on diffraction about the head with only two ears. In this paper we present an integrated robot phonotaxis system which utilizes the robot's movement to resolve frontback localization ambiguity. Our system achieves high angular localization acuity (/spl plusmn/ 2/spl deg/) and it was successfully tested in localizing a single broadband source and moving towards it within a cluttered laboratory environment.

We propose a landmark-based representation of maps to be used for robot navigation and exploration. Our approach is aimed towards mobile robots that operate over expansive, imprecisely known terrain without a single "global" map. Instead, a map is pieced together from local terrain and navigation data stored in a directed graph. Each of the graph's vertices contains information describing a landmark locally (e.g. a detailed map of that landmark's immediate surroundings). The geometric relationships between landmarks are unknown. Graph edges store language-based directions that enable a robot to steer between landmarks. These directions are written in the motion description language MDLe, reducing the complexity of the map and making navigation programs robot-independent. Furthermore, the proposed architecture is economical with respect to the amount of storage required to describe far-flung areas of interest. We present preliminary results demonstrating our ideas using an indoor robot.