An official website of the United States government
In this paper we propose a landmark-based map localization system for robotic swarms. The proposed system leverages the capabilities of a distributed landmark identification algorithm developed for robotic swarms presented in [1]. The output of the landmark identification consists of a vector of probabilities that each individual robot is looking at a particular landmark in the environment. In this work, this vector is used individually by each component of the swarm to feed the measurement update of a particle filter to estimate the robot location. The system was tested in simulation to validate its performance.
more »
« less
Cooperative place recognition in robotic swarms
In this paper we propose a study on landmark identification as a step towards a localization setup for real-world robotic swarms setup. In real world, landmark identification is often tackled as a place recognition problem through the use of computationally intensive Convolutional Neural Networks. However, the components of a robotic swarm usually have limited computational and sensing capabilities that allows only for the application of relatively shallow networks that results in large percentage of recognition errors. In a previous attempt of solving a similar setup - cooperative object recognition - the authors of [1] have demonstrated how the use of communication among a swarm and a naive Bayes classifier was able to substantially improve the correct recognition rate. An assumption of that paper not compatible with a swarm localization setup was that all swarm components would be looking at the same object. In this paper, we propose the use of a weighting factor to relapse this assumption. Through the use of simulation data, we show that our approach provides high recognition rates even in situations in which the robots would look at different objects.
more »
« less
- Award ID(s):
- 1952862
- PAR ID:
- 10345831
- Date Published:
- Journal Name:
- Proceedings of the 36th Annual ACM Symposium on Applied Computing
- Page Range / eLocation ID:
- 785 to 792
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
This paper presents a formulation for swarm control and high-level task planning that is dynamically responsive to user commands and adaptable to environmental changes. We design an end-to-end pipeline from a tactile tablet interface for user commands to onboard control of robotic agents based on decentralized ergodic coverage. Our approach demonstrates reliable and dynamic control of a swarm collective through the use of ergodic specifications for planning and executing agent trajectories as well as responding to user and external inputs. We validate our approach in a virtual reality simulation environment objectives in real-time. and in real-world experiments at the DARPA OFFSET Urban Swarm Challenge FX3 field tests with a robotic swarm where user-based control of the swarm and mission-based tasks require a dynamic and flexible response to changing conditions and objectives in real-time.more » « less
-
This paper outlines a methodology for the construction of vector fields that can enable a multi-robot system moving on the plane to generate multiple dynamical behaviors by adjusting a single scalar parameter. This parameter essentially triggers a Hopf bifurcation in an underlying time-varying dynamical system that steers a robotic swarm. This way, the swarm can exhibit a variety of behaviors that arise from the same set of continuous differential equations. Other approaches to bifurcation-based swarm coordination rely on agent interaction which cannot be realized if the swarm members cannot sense or communicate with one another. The contribution of this paper is to offer an alternative method for steering minimally instrumented multi-robot collectives with a control strategy that can realize a multitude of dynamical behaviors without switching their constituent equations. Through this approach, analytical solutions for the bifurcation parameter are provided, even for more complex cases that are described in the literature, along with the process to apply this theory in a multi-agent setup. The theoretical predictions are confirmed via simulation and experimental results with the latter also demonstrating real-world applicability.more » « less
-
This work addresses the problem of Social Activity Recognition (SAR), a critical component in real-world tasks like surveillance and assistive robotics. Unlike traditional event understanding approaches, SAR necessitates modeling individual actors' appearance and motions and contextualizing them within their social interactions. Traditional action localization methods fall short due to their single-actor, single-action assumption. Previous SAR research has relied heavily on densely annotated data, but privacy concerns limit their applicability in real-world settings. In this work, we propose a self-supervised approach based on multi-actor predictive learning for SAR in streaming videos. Using a visual-semantic graph structure, we model social interactions, enabling relational reasoning for robust performance with minimal labeled data. The proposed framework achieves competitive performance on standard group activity recognition benchmarks. Evaluation on three publicly available action localization benchmarks demonstrates its generalizability to arbitrary action localization.more » « less
-
Training robotic policies in simulation suffers from the sim-to-real gap, as simulated dynamics can be different from real-world dynamics. Past works tackled this problem through domain randomization and online system-identification. The former is sensitive to the manually-specified training distribution of dynamics parameters and can result in behaviors that are overly conservative. The latter requires learning policies that concurrently perform the task and generate useful trajectories for system identification. In this work, we propose and analyze a framework for learning exploration policies that explicitly perform task-oriented exploration actions to identify task-relevant system parameters. These parameters are then used by model-based trajectory optimization algorithms to perform the task in the real world. We instantiate the framework in simulation with the Linear Quadratic Regulator as well as in the real world with pouring and object dragging tasks. Experiments show that task-oriented exploration helps model-based policies adapt to systems with initially unknown parameters, and it leads to better task performance than task-agnostic exploration.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3412841.3441954
