Information in bicycle pelotons consists of two main types: displayed information that is perceptible to others; and hidden information available to individual riders about their own physical state. Flow (or transfer) of information in pelotons occurs in two basic ways: 1) between cyclists within a peloton, which riders exploit to adjust tactical objectives (“intra-peloton”); 2) from sources outside a peloton as it is fed to riders via radio communication, or from third parties (“extra-peloton”). A conceptual framework is established for information transfer intra-peloton and extra-peloton. Both kinds of information transfer affect peloton complex dynamics. Pelotons exhibit mixed self-organized and top-down dynamics. These can be isolated and examined independently: self-organized dynamics emerge through local physical rules of interaction, and are distinguishable from the top-down dynamics of human competition, decision-making and information transfer. Both intra and extra-peloton information flow affect individual rider positions and the timing of their positional changes, but neither types of peloton information flow fundamentally alter self-organized structures. In addition to two previously identified peloton resources for which riders compete - energy saved by drafting, and near-front positions - information flow is identified as a third peloton resource. Also, building upon previous work on peloton phase-transitions and self-organized group-sorting, identified here is a transition between a team cluster state in which team-mates ride near each other, and a self-organized “fitness” cluster state in which riders of near equal fitness levels gravitate toward each other.

This paper presents a framework that facilitates communication between a planning system (“planner”) and a plan execution system (“executor”) to enable them to run concurrently, with the main emphasis on meeting the real-time requirements of the application domain. While the framework is applicable to general-purpose planning, its features are optimized for the requirements of automated story planning for games—with emphasis on monitoring player-triggered events and handling on-time (re-)generation of story assets such as characters, maps and scenarios. This framework subsumes the traditional interleaved planning-and-execution paradigm used in embedded continual planning systems and generalizes it to a non-embedded context, making the framework ideal for use with contemporary game architectures (e.g., multithreaded game engines, or games with subsystems communicating over a network).

Imitation can be viewed as a means of enhancing learning in multiagent environments. It augments an agent's ability to learn useful behaviors by making intelligent use of the knowledge implicit in behaviors demonstrated by cooperative teachers or other more experienced agents. We propose and study a formal model of implicit imitation that can accelerate reinforcement learning dramatically in certain cases. Roughly, by observing a mentor, a reinforcement-learning agent can extract information about its own capabilities in, and the relative value of, unvisited parts of the state space. We study two specific instantiations of this model, one in which the learning agent and the mentor have identical abilities, and one designed to deal with agents and mentors with different action sets. We illustrate the benefits of implicit imitation by integrating it with prioritized sweeping, and demonstrating improved performance and convergence through observation of single and multiple mentors. Though we make some stringent assumptions regarding observability and possible interactions, we briefly comment on extensions of the model that relax these restricitions.

One of the challenges in accurately applying metrics for life cycle assessment lies in accounting for both irreducible and inherent uncertainties in how a design will perform under real world conditions. This paper presents a preliminary study that compares two strategies, one simulation-based and one set-based, for propagating uncertainty in a system. These strategies for uncertainty propagation are then aggregated. This work is conducted in the context of an amorphous photovoltaic (PV) panel, using data gathered from the National Solar Radiation Database, as well as realistic data collected from an experimental hardware setup specifically for this study. Results show that the influence of various sources of uncertainty can vary widely, and in particular that solar radiation intensity is a more significant source of uncertainty than the efficiency of a PV panel. This work also shows both set-based and simulation-based approaches have limitations and must be applied thoughtfully to prevent unrealistic results. Finally, it was found that aggregation of the two uncertainty propagation methods provided faster results than either method alone.

A peloton is a group of cyclists whose individual and collective energy expenditures are reduced when cyclists ride behind others in zones of reduced air pressure; this effect is known in cycling as ‘drafting’. Through drafting cyclists couple their energy expenditures. Coupling of cyclists’ energy expenditures when drafting is the basic peloton property from which self-organized collective behaviours emerge. Here we examine peloton hysteresis, applying the definition used in the context of vehicle traffic in which a rapid deceleration to a high density state (jam) is followed by a lag in vehicle acceleration. Applying a flow analysis of volume (number of cyclists) over time, peloton hysteresis is examined in three forms: one is similar to vehicle traffic hysteresis in which rapid decelerations and increased flow (or density) are followed by extended acceleration periods and reduced flow. In cycling this is known as the accordion effect. A second kind of hysteresis results from rapid accelerations followed by periods of decreasing speeds and decreasing flow. This form of hysteresis is essentially inverse to traffic hysteresis and the accordion effect. We show this form of hysteresis using data from a mass-start bicycle points-race. A third kind of peloton hysteresis occurs when the drafting benefit is minimized on hills and weaker cyclists lose positions in the peloton, while flow/density is retained. A computer simulation shows this hysteresis among two sets of cyclist agents, each with different output capacity and models hysteresis as a peloton transitions from flat topography to a steep incline on which drafting is negligible.

A wide range of problems, from contingent and multiagent planning to process/service orchestration, can be viewed as games. In many of these, it is natural to spec- ify the possible behaviors procedurally. In this paper, we develop a logical framework for specifying these types of problems/games based on the situation calculus and ConGolog. The framework incorporates game-theoretic path quantifiers as in ATL. We show that the framework can be used to model such problems in a natural way. We also show how verification/synthesis techniques can be used to solve problems expressed in the framework. In particular, we develop a method for dealing with infinite state settings using fixpoint approximation and “characteristic graphs”.

A peloton is a group of cyclists whose individual and collective energy expenditures are reduced when cyclists ride behind others in zones of reduced air pressure; this effect is known in cycling as ‘drafting’. As an aggregate of biological agents (human), a peloton is a complex dynamical system from which patterns of collective behaviour emerge, including phases and transitions between phases, through which pelotons oscillate. Coupling of cyclists’ energy expenditures when drafting is the basic peloton property from which self-organized collective behaviours emerge. Shown here are equations that model coupling behaviours. Environmental constraints are further parameters that affect peloton dynamics. Phases are defined by thresholds of aggregate energy expenditure; shown here are two different, but consistent, conceptual descriptions of these phase transitions. The first is an energetic model that describes phases in terms of individual, bi-coupled and globally-coupled energy output thresholds that define four observable changes in peloton behaviour. A second, economic model incorporates competition and cooperation dynamics: cooperation increases as power outputs and course constraints increase and population diminishes, and where competition and cooperation for resources results in peloton divisions into sub-pelotons whose average fitness levels are more closely homogeneous.

Standard value function approaches to finding policies for Partially Observable Markov Decision Processes (POMDPs) are generally considered to be intractable for large models. The intractability of these algorithms is to a large extent a consequence of computing an exact, optimal policy over the entire belief space. However, in real-world POMDP problems, computing the optimal policy for the full belief space is often unnecessary for good control even for problems with complicated policy classes. The beliefs experienced by the controller often lie near a structured, low-dimensional subspace embedded in the high-dimensional belief space. Finding a good approximation to the optimal value function for only this subspace can be much easier than computing the full value function. We introduce a new method for solving large-scale POMDPs by reducing the dimensionality of the belief space. We use Exponential family Principal Components Analysis (Collins, Dasgupta & Schapire, 2002) to represent sparse, high-dimensional belief spaces using small sets of learned features of the belief state. We then plan only in terms of the low-dimensional belief features. By planning in this low-dimensional space, we can find policies for POMDP models that are orders of magnitude larger than models that can be handled by conventional techniques. We demonstrate the use of this algorithm on a synthetic problem and on mobile robot navigation tasks.

Imitation can be viewed as a means of enhancing learning in multiagent environments. It augments an agent's ability to learn useful behaviors by making intelligent use of the knowledge implicit in behaviors demonstrated by cooperative teachers or other more experienced agents. We propose and study a formal model of implicit imitation that can accelerate reinforcement learning dramatically in certain cases. Roughly, by observing a mentor, a reinforcement-learning agent can extract information about its own capabilities in, and the relative value of, unvisited parts of the state space. We study two specific instantiations of this model, one in which the learning agent and the mentor have identical abilities, and one designed to deal with agents and mentors with different action sets. We illustrate the benefits of implicit imitation by integrating it with prioritized sweeping, and demonstrating improved performance and convergence through observation of single and multiple mentors. Though we make some stringent assumptions regarding observability and possible interactions, we briefly comment on extensions of the model that relax these restricitions.

There is an increasing need for automated support for humans monitoring the activity of distributed teams of cooperating agents, both human and machine. We characterize the domain-independent challenges posed by this problem, and describe how properties of domains influence the challenges and their solutions. We will concentrate on dynamic, data-rich domains where humans are ultimately responsible for team behavior. Thus, the automated aid should interactively support effective and timely decision making by the human. We present a domain-independent categorization of the types of alerts a plan-based monitoring system might issue to a user, where each type generally requires different monitoring techniques. We describe a monitoring framework for integrating many domain-specific and task-specific monitoring techniques and then using the concept of value of an alert to avoid operator overload. We use this framework to describe an execution monitoring approach we have used to implement Execution Assistants (EAs) in two different dynamic, data-rich, real-world domains to assist a human in monitoring team behavior. One domain (Army small unit operations) has hundreds of mobile, geographically distributed agents, a combination of humans, robots, and vehicles. The other domain (teams of unmanned ground and air vehicles) has a handful of cooperating robots. Both domains involve unpredictable adversaries in the vicinity. Our approach customizes monitoring behavior for each specific task, plan, and situation, as well as for user preferences. Our EAs alert the human controller when reported events threaten plan execution or physically threaten team members. Alerts were generated in a timely manner without inundating the user with too many alerts (less than 10 percent of alerts are unwanted, as judged by domain experts).