Active information acquisition is an optimal control problem aiming to minimize an uncertainty measure over the Bayesian dynamics of a controlled estimation process. Examples of problems that can be formulated as active information acquisition include active simultaneous localization and mapping, in which the robots autonomously explore and map an unknown environment, and target tracking, in which a team of robots aims to detect, localize, and track multiple moving targets. Our work considers the design of control policies for n robots that aim to minimize the uncertainty (e.g., entropy) in the joint target-robot state over a planning horizon of length T. We are developing techniques for Bayesian inference and uncertainty reducing motion planning that scale efficiently in both the horizon length, T, and the number of robots, n.
Focusing on the complexity in T, we show that active information acquisition with linear Gaussian models of the target dynamics and observations reduces from a stochastic to a deterministic optimal control problem over the space of sensor states and target covariance matrices. In this linear Gaussian case, the optimal estimator is the Kalman filter, the entropy is proportional to the log-determinant of the target covariance matrix, and the monotonicity and concavity of the Riccati map allow us to design efficient (in T) planning algorithms. Our algorithms exploit these properties to obtain bounds on the long-term cost of different states allowing branch-and-bound and heuristic function techniques and theoretical guarantees of bounded suboptimality regardless of the length of the planning horizon. Our algorithms allow an adjustable trade-off between planning time and solution quality to efficiently solve information gathering problems with long planning horizons. In practical settings, where the target motion and observation models are generally nonlinear and non-Gaussian, approximate Gaussian inference and repeated replanning using our algorithm can generate a closed-loop sensing policy.
Focusing on the complexity in n, we design distributed estimation and planning algorithms that rely only on local neighborhood communication with robot teammates. Our estimation work develops algorithms, supported by theoretical performance guarantees, for relative localization (estimating robot states using relative measurements of the neighboring robot states) and distributed Bayesian inference (estimating a common phenomenon of interest, such as a map of the environment or the motion of multiple targets, by sharing sensing information across the robot network). We propose Gaussian Mixture Multi Dimensional Scaling (GM-MDS) and barycentric-coordinate algorithms for range-only localization, e.g., using ultra-wideband radio measurements. More generally, we develop distributed Bayesian inference techniques, in which each robot maintains and updates its own probability density function (pdf) over the target states and averages it with the pdfs of its one-hop communication neighbors to achieve consensus to a globally consistent target pdf. Our work establishes that such distributed probabilistic estimation techniques are consistent (e.g., almost surely or in mean square) under mild requirements on the robot team topology (always eventually strongly connected). Our work on distributed planning for active estimation uncertainty reduction exploits the submodularity of information measures. The measurements obtained along the trajectory of an additional robot provide diminishing returns with respect to the (mutual) information collected about the targets. The setting can be formalized as a partition matroid over the space of robot trajectories. We prove that decentralized planning schemes that scale linearly with the number of robots, n, provide guaranteed performance of at least 50% of the (mutual) information achievable by the optimal centralized sensing plan.
@inproceedings{Atanasov_ActiveInformationAcquisition_ICRA15,
author = {N. Atanasov and J. Le Ny and K. Daniilidis and G. Pappas},
title = {Decentralized Active Information Acquisition: Theory and Application to Multi-Robot SLAM},
booktitle = {IEEE Int. Conf. on Robotics and Automation (ICRA)},
year = {2015},
pages = {4775-4782},
doi = {http://www.doi.org/10.1109/ICRA.2015.7139863}
}
@inproceedings{Atanasov_ActiveInformationAcquisition_ICRA14,
author = {N. Atanasov and J. Le Ny and K. Daniilidis and G. Pappas},
title = {Information Acquisition with Sensing Robots: Algorithms and Error Bounds},
booktitle = {IEEE Int. Conf. on Robotics and Automation (ICRA)},
year = {2014},
pages = {6447-6454},
doi = {http://www.doi.org/10.1109/ICRA.2014.6907811}
}
@inproceedings{Atanasov_JointEstimationLocalization_CDC14,
author = {N. Atanasov and R. Tron and V. Preciado and G. Pappas},
title = {Joint Estimation and Localization in Sensor Networks},
booktitle = {IEEE Conf. on Decision and Control (CDC)},
year = {2014},
pages = {6875-6882},
doi = {http://www.doi.org/10.1109/CDC.2014.7040469}
}
@article{Schlotfeldt_AnytimeInfoGathering_RAL18,
author = {Schlotfeldt, Brent and Thakur, Dinesh and Atanasov, Nikolay and Kumar, Vijay and Pappas, George J.},
title = {Anytime Planning for Decentralized Multi-Robot Active Information Gathering},
journal = {IEEE Robotics and Automation Letters (RAL)},
year = {2018},
volume = {3},
number = {2},
pages = {1025-1032},
doi = {http://www.doi.org/10.1109/LRA.2018.2794608}
}
@inproceedings{Schlotfeldt_InfoBounds_IROS19,
author = {B. Schlotfeldt and N. Atanasov and G. J. Pappas},
title = {Maximum Information Bounds for Planning Active Sensing Trajectories},
booktitle = {IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS)},
year = {2019},
doi = {https://www.doi.org/10.1109/IROS40897.2019.8968147}
}
@inproceedings{Kantaros_InformationGathering_RSS19,
author = {Y. Kantaros and B. Schlotfeldt and N. Atanasov and G. J. Pappas},
title = {Asymptotically Optimal Planning for Non-Myopic Multi-Robot Information Gathering},
booktitle = {Robotics: Science and Systems (RSS)},
year = {2019},
doi = {https://www.doi.org/10.15607/RSS.2019.XV.062}
}
@inproceedings{Ostertag_MinimumUncertaintyControl_ACC19,
author = {M. Ostertag and N. Atanasov and T. Rosing},
title = {Robust Velocity Control for Minimum Steady State Uncertainty in Persistent Monitoring Applications},
booktitle = {American Control Conference (ACC)},
year = {2019},
doi = {https://www.doi.org/10.23919/ACC.2019.8814376}
}
@inproceedings{Tzoumas_SensorSchedululing_ACC17,
author = {V. Tzoumas and N. Atanasov and A. Jadbabaie and G. Pappas},
title = {Scheduling Nonlinear Sensors for Stochastic Process Estimation},
booktitle = {American Control Conference (ACC)},
year = {2017},
doi = {http://www.doi.org/10.23919/ACC.2017.7963015}
}
@inproceedings{Paritosh_MarginalDensityAveraging_CDC20,
author = {P. Paritosh and N. Atanasov and S. Martinez},
title = {Marginal Density Averaging for Distributed Node Localization from Local Edge Measurements},
booktitle = {IEEE Conference on Decision and Control (CDC)},
year = {2020}
}
@inproceedings{Paritosh_DistributedEstimation_CDC19,
author = {P. Paritosh and N. Atanasov and S. Martinez},
title = {Hypothesis Assignment and Partial Likelihood Averaging for Cooperative Estimation},
booktitle = {IEEE Conference on Decision and Control (CDC)},
year = {2019},
doi = {https://www.doi.org/10.1109/CDC40024.2019.9029768}
}
@inproceedings{Tecchio_RangeLocalization_ACC19,
author = {P. Tecchio and N. Atanasov and S. Shahrampour and G. J. Pappas},
title = {N-Dimensional Distributed Network Localization With Noisy Range Measurements and Arbitrary Anchor Placement},
booktitle = {American Control Conference (ACC)},
year = {2019},
doi = {https://www.doi.org/10.23919/ACC.2019.8814820}
}
@inproceedings{DiFranco_UWBNetworkLocalization_IPSN17,
author = {C. Di Franco and A. Prorok and N. Atanasov and B. Kempke and P. Dutta and V. Kumar and G. Pappas},
title = {Calibration-Free Network Localization using Non-Line-of-Sight Ultra-Wideband Measurements},
booktitle = {ACM/IEEE Int. Conf. on Information Processing in Sensor Networks (IPSN)},
year = {2017},
doi = {http://www.doi.org/10.1145/3055031.3055091}
}
@article{Atanasov_StochasticSourceSeeking_JDSMC14,
author = {N. Atanasov and J. Le Ny and G. Pappas},
title = {Distributed Algorithms for Stochastic Source Seeking with Mobile Robot Networks},
journal = {ASME Journal of Dynamic Systems, Measurement, and Control (JDSMC)},
year = {2015},
volume = {137},
number = {3},
pages = {031011-031011-9},
doi = {http://www.doi.org/10.1115/1.4027892}
}
@inproceedings{Schlotfeldt_AdversarialInfoAcquisition_RSS18_Workshop,
author = {B. Schlotfeldt and N. Atanasov and G. J. Pappas},
title = {Adversarial Information Acquisition},
booktitle = {Workshop on Adversarial Robotics at RSS},
year = {2018}
}