Charles Pippin

Cost based planning with RRT in outdoor environments

The Rapidly Exploring Random Tree (RRT) algorithm can be applied to the robotic path planning problem and performs well in challenging, dynamic domains. Traditional RRT methods use a binary cost function and they select portions of the tree for expansion based on the Euclidean distance to the target. However, in outdoor navigation, the relative cost of terrain can also provide useful input to a planning algorithm that traditional RRT methods cannot take advantage of. We present the Metric Adaptive RRT (MA-RRT), which integrates planning and fast execution for generating paths over a cost map. The MA-RRT algorithm considers underlying cost of a path when calculating the distance function for tree expansion. A heuristic value is also used for determining distance from a point to the target and an adaptive mechanism is employed for adjusting the heuristic on-line. We have implemented our approach in offline simulations and in outdoor robot experiments, and show that the MA-RRT algorithm can improve upon the quality of the path returned when cost is considered. The trade off between cost consideration and runtime performance is also presented.

Performance based task assignment in multi-robot patrolling

This article applies a performance metric to the multi-robot patrolling task to more efficiently distribute patrol areas among robot team members. The multi-robot patrolling task employs multiple robots to perform frequent visits to known areas in an environment, while minimizing the time between node visits. Conventional strategies for performing this task assume that the robots will perform as expected and do not address situations in which some team members patrol inefficiently. However, reliable performance of team members may not always be a valid assumption. This paper considers an approach for monitoring robot performance in a patrolling task and dynamically reassigning tasks from those team members that perform poorly. Experimental results from simulation and on a team of indoor robots demonstrate that in using this approach, tasks can be dynamically and more efficiently distributed in a multi-robot patrolling application.

Finding Optimal Routes for Multi-Robot Patrolling in Generic Graphs

Multi-robot patrolling is a problem that has important applications in security and surveillance. However, the optimal task assignment is known to be NP-hard. We consider evenly spacing the robots in a cyclic Traveling Salesman Problem (TSP) tour or partitioning the graph of the environment. The trade-off in performance, overall team travel cost and coordination is analyzed in this paper. We provide both a theoretical analysis and simulation results across multiple environments. The results demonstrate that generally cyclic-based strategies are superior, especially when small teams are used but at the expense of greater team cost, whereas partitioning strategies are especially suitable for larger teams and unbalanced graph topologies. The reported results show that graph topology and team size are fundamental to determine the best choice for a patrol strategy.

Trust Modeling in Multi-Robot Patrolling

On typical multi-robot teams, there is an implicit assumption that robots can be trusted to effectively perform assigned tasks. The multi-robot patrolling task is an example of a domain that is particularly sensitive to reliability and performance of robots. Yet reliable performance of team members may not always be a valid assumption even within homogeneous teams. For instance, a robots performance may deteriorate over time or a robot may not estimate tasks correctly. Robots that can identify poorly performing team members as performance deteriorates, can dynamically adjust the task assignment strategy. This paper investigates the use of an observation based trust model for detecting unreliable robot team members. Robots can reason over this model to perform dynamic task reassignment to trusted team members. Experiments were performed in simulation and using a team of indoor robots in a patrolling task to demonstrate both centralized and decentralized approaches to task reassignment. The results clearly demonstrate that the use of a trust model can improve performance in the multi-robot patrolling task.

Bio-inspired multi-robot communication through behavior recognition

This paper focuses on enabling multi-robot teams to cooperatively perform tasks without the use of radio or acoustic communication. One key to more effective cooperative interaction in a multi-robot team is the ability to understand the behavior and intent of other robots. This is similar to the honey bee “waggle dance” in which a bee can communicate the orientation and distance of a food source. In this similar manner, our heterogenous multi-robot team uses a specific behavior to indicate the location of mine-like objects (MLOs). Observed teammate action sequences can be learned to perform behavior recognition and task-assignment in the absence of communication. We apply Conditional Random Fields (CRFs) to perform behavior recognition as an approach to task monitoring in the absence of communication in a challenging underwater environment. In order to demonstrate the use of behavior recognition of an Autonomous Underwater Vehicle (AUV) in a cooperative task, we use trajectory based techniques for model generation and behavior discrimination in experiments using simulated scenario data. Results are presented demonstrating heterogenous teammate cooperation between an AUV and an Autonomous Surface Vehicle (ASV) using behavior recognition rather than radio or acoustic communication in a mine clearing task.

A Bayesian formulation for auction-based task allocation in heterogeneous multi-agent teams

In distributed, heterogeneous, multi-agent teams, agents may have different capabilities and types of sensors. Agents in dynamic environments will need to cooperate in real-time to perform tasks with minimal costs. Some example scenarios include dynamic allocation of UAV and UGV robot teams to possible hurricane survivor locations, search and rescue and target detection. Auction based algorithms scale well because agents generally only need to communicate bid information. In addition, the agents are able to perform their computations in parallel and can operate on local information. Furthermore, it is easy to integrate humans and other vehicle types and sensor combinations into an auction framework. However, standard auction mechanisms do not explicitly consider sensors with varying reliability. The agents sensor qualities should be explicitly accounted. Consider a scenario with multiple agents, each carrying a single sensor. The tasks in this case are to simply visit a location and detect a target. The sensors are of varying quality, with some having a higher probability of target detection. The agents themselves may have different capabilities, as well. The agents use knowledge of their environment to submit cost-based bids for performing each task and an auction is used to perform the task allocation. This paper discusses techniques for including a Bayesian formulation of target detection likelihood into this auction based framework for performing task allocation across multi-agent heterogeneous teams. Analysis and results of experiments with multiple air systems performing distributed target detection are also included.

Dynamic, cooperative multi-robot patrolling with a team of UAVs

The multi-robot patrolling task has practical relevance in surveillance, search and rescue, and security appli- cations. In this task, a team of robots must repeatedly visit areas in the environment, minimizing the time in-between visits to each. A team of robots can perform this task efficiently; however, challenges remain related to team formation and task assignment. This paper presents an approach for monitoring patrolling performance and dynamically adjusting the task assignment function based on observations of teammate performance. Experimental results are presented from realistic simulations of a cooperative patrolling scenario, using a team of UAVs.

AUV behavior recognition using behavior histograms, HMMs, and CRFs

This paper focuses on behavior recognition in an underwater application as a substitute for communicating through acoustic transmissions, which can be unreliable. The importance of this work is that sensor information regarding other agents can be leveraged to perform behavior recognition, which is activity recognition of robots performing specific programmed behaviors, and task-assignment. This work illustrates the use of Behavior Histograms, Hidden Markov Models (HMMs), and Conditional Random Fields (CRFs) to perform behavior recognition. We present challenges associated with using each behavior recognition technique along with results on individually selected test trajectories, from simulated and real sonar data, and real-time recognition through a simulated mission.

Cooperation based Dynamic Team Formation in Multi-Agent Auctions

Auction based methods are often used to perform distributed task allocation on multi-agent teams. Many existing approaches to auctions assume fully cooperative team members. On in-situ and dynamically formed teams, reciprocal collaboration may not always be a valid assumption. This paper presents an approach for dynamically selecting auction partners based on observed team member performance and shared reputation. In addition, we present the use of a shared reputation authority mechanism. Finally, experiments are performed in simulation on multiple UAV platforms to highlight situations in which it is better to enforce cooperation in auctions using this approach.

Learning Task Performance in Market-Based Task Allocation

Auction based algorithms offer effective methods for de-centralized task assignment in multi-agent teams. Typically there is an implicit assumption that agents can be trusted to effectively perform assigned tasks. However, reliable performance of team members may not always be a valid assumption. An approach to learning team member performance is presented, which enables more efficient task assignment. A policy gradient reinforcement learning algorithm is used to learn a cost factor that can be applied individually to auction bids. Experimental results demonstrate that agents that model team member performance using this approach can more efficiently distribute tasks in multi-agent auctions.