Brian D. Ziebart

Activity forecasting

We address the task of inferring the future actions of people from noisy visual input. We denote this task activity forecasting. To achieve accurate activity forecasting, our approach models the effect of the physical environment on the choice of human actions. This is accomplished by the use of state-of-the-art semantic scene understanding combined with ideas from optimal control theory. Our unified model also integrates several other key elements of activity analysis, namely, destination forecasting, sequence smoothing and transfer learning. As proof-of-concept, we focus on the domain of trajectory-based activity analysis from visual input. Experimental results demonstrate that our model accurately predicts distributions over future actions of individuals. We show how the same techniques can improve the results of tracking algorithms by leveraging information about likely goals and trajectories.

Planning-based prediction for pedestrians

We present a novel approach for determining robot movements that efficiently accomplish the robots tasks while not hindering the movements of people within the environment. Our approach models the goal-directed trajectories of pedestrians using maximum entropy inverse optimal control. The advantage of this modeling approach is the generality of its learned cost function to changes in the environment and to entirely different environments. We employ the predictions of this model of pedestrian trajectories in a novel incremental planner and quantitatively show the improvement in hindrance-sensitive robot trajectory planning provided by our approach.

Navigate like a cabbie: probabilistic reasoning from observed context-aware behavior

We present PROCAB, an efficient method for Probabilistically Reasoning from Observed Context-Aware Behavior. It models the context-dependent utilities and underlying reasons that people take different actions. The model generalizes to unseen situations and scales to incorporate rich contextual information. We train our model using the route preferences of 25 taxi drivers demonstrated in over 100,000 miles of collected data, and demonstrate the performance of our model by inferring: (1) decision at next intersection, (2) route to known destination, and (3) destination given partially traveled route.

Towards a pervasive computing benchmark

Pervasive computing allows the coupling of the physical world to the information world, and provides a wealth of ubiquitous services and applications that allow users, machines, data, applications, and physical spaces to interact seamlessly with one another. In this paper, we propose a benchmark for evaluating pervasive computing environments. These proposed metrics facilitate the assessment and evaluation of different aspects of pervasive computing and its support for a wide variety of tasks.

Learning patterns of pick-ups and drop-offs to support busy family coordination

Part of being a parent is taking responsibility for arranging and supplying transportation of children between various events. Dual-income parents frequently develop routines to help manage transportation with a minimal amount of attention. On days when families deviate from their routines, effective logistics can often depend on knowledge of the routine location, availability and intentions of other family members. Since most families rarely document their routine activities, making that needed information unavailable, coordination breakdowns are much more likely to occur. To address this problem we demonstrate the feasibility of learning family routines using mobile phone GPS. We describe how we (1) detect pick-ups and drop-offs; (2) predict which parent will perform a future pick-up or drop-off; and (3) infer if a child will be left at an activity. We discuss how these routine models give digital calendars, reminder and location systems new capabilities to help prevent breakdowns, and improve family life.

The Principle of Maximum Causal Entropy for Estimating Interacting Processes

The principle of maximum entropy provides a powerful framework for estimating joint, conditional, and marginal probability distributions. However, there are many important distributions with elements of interaction and feedback where its applicability has not been established. This paper presents the principle of maximum causal entropy-an approach based on directed information theory for estimating an unknown process based on its interactions with a known process. We demonstrate the breadth of the approach using two applications: a predictive solution for inverse optimal control in decision processes and computing equilibrium strategies in sequential games.

Both nearest neighbours and long-term affiliates predict individual locations during collective movement in wild baboons

In many animal societies, groups of individuals form stable social units that are shaped by well-delineated dominance hierarchies and a range of affiliative relationships. How do socially complex groups maintain cohesion and achieve collective movement? Using high-resolution GPS tracking of members of a wild baboon troop, we test whether collective movement in stable social groups is governed by interactions among local neighbours (commonly found in groups with largely anonymous memberships), social affiliates, and/or by individuals paying attention to global group structure. We construct candidate movement prediction models and evaluate their ability to predict the future trajectory of focal individuals. We find that baboon movements are best predicted by 4 to 6 neighbours. While these are generally individuals’ nearest neighbours, we find that baboons have distinct preferences for particular neighbours, and that these social affiliates best predict individual location at longer time scales (>10 minutes). Our results support existing theoretical and empirical studies highlighting the importance of local rules in driving collective outcomes, such as collective departures, in primates. We extend previous studies by elucidating the rules that maintain cohesion in baboons ‘on the move’, as well as the different temporal scales of social interactions that are at play.

Leveraging Machine Learning to Improve Unwanted Resource Filtering

Advertisements simultaneously provide both economic support for most free web content and one of the largest annoyances to end users. Furthermore, the modern advertisement ecosystem is rife with tracking methods which violate user privacy. A natural reaction is for users to install ad blockers which prevent advertisers from tracking users or displaying ads. Traditional ad blocking software relies upon hand-crafted filter expressions to generate large, unwieldy regular expressions matched against resources being included within web pages. This process requires a large amount of human overhead and is susceptible to inferior filter generation. We propose an alternate approach which leverages machine learning to bootstrap a superior classifier for ad blocking with less human intervention. We show that our classifier can simultaneously maintain an accuracy similar to the hand-crafted filters while also blocking new ads which would otherwise necessitate further human intervention in the form of additional handmade filter rules.

Goal-predictive robotic teleoperation from noisy sensors

Robotic teleoperation from a human operators pose demonstrations provides an intuitive and effective means of control that has been made feasible by improvements in sensor technologies in recent years. However, the imprecision of low-cost depth cameras and the difficulty of calibrating a frame of reference for the operator introduce inefficiencies in this process when performing tasks that require interactions with objects in the robots workspace. We develop a goal-predictive teleoperation system that aids in “de-noising” the controls of the operator to be more goal-directed. Our approach uses inverse optimal control to predict the intended object of interaction from the current motion trajectory in real time and then adapts the degree of autonomy between the operators demonstrations and autonomous completion of the predicted task. We evaluate our approach using the Microsoft Kinect depth camera as our input sensor to control a Rethink Robotics Baxter robot.

3 rd workshop on recent advances in behavior prediction and pro-active pervasive computing

The 2nd Workshop on recent advances in behavior prediction and pro-active pervasive computing focuses on contributions that target recent challenges of context prediction and on applications of context prediction. The main challenges are a lack of benchmarks and common data sets, as well as a lack of development frameworks and that the main focus of context prediction still remains location prediction. Since context prediction is a key requirement to enable proactive applications, the workshop aims to intensify the discussion about the state and direction of context prediction research and to facilitate collaboration among research groups focusing on context prediction.