Contextual Sequence Prediction with Application to Control Library Optimization

By Debadeepta Dey, Tian Yu Tommy Liu, Martial Hebert, J. Bagnell. 2012

45 citations recorded by the source corpus.

Sequence optimization, where the items in a list are ordered to maximize some reward has many applications such as web advertisement placement, search, and control libraries in robotics. Previous work in sequence optimization produces a static ordering that does not take any features of the item or context of the problem into account. In this work, we propose a general approach to order the items within the sequence based on the context (e.g., perceptual information, environment description, and goals). We take a simple, efficient, reduction-based approach where the choice and order of the items is established by repeatedly learning simple classifiers or regressors for each “slot” in the sequence. Our approach leverages recent work on submodular function maximization to provide a formal regret reduction from submodular sequence optimization to simple costsensitive prediction. We apply our contextual sequence prediction algorithm to optimize control libraries and demonstrate results on two robotics problems: manipulator trajectory prediction and mobile robot path planning.

Subfields: Computer Vision and Pattern Recognition, Model Architectures, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Model Architectures

In Signal Processing

Citations in the corpus

Learning Trajectory Preferences for Manipulators via Iterative Improvement (2013, background): “However, for high DoF manipulators sampling based approaches [4, 6] maintains tractability of the problem, hence we take this approach.”
Legibility and predictability of robot motion (2013, background): “While efficient trajectory optimization techniques do exist for high-dimensional spaces and non-convex costs [34], they are subject to local minima, and how to alleviate this issue in practice remains an open research question [35], [36].”
CHOMP: Covariant Hamiltonian optimization for motion planning (2013, background): “5, which makes CHOMP probabilistically complete), or through learning from previous experiences to initialize the optimizer in a good basin of attraction [16, 20].”
Why M Heads are Better than One: Training a Diverse Ensemble of Deep Networks (2015, background): “[6] in the context of submodular list prediction.”
A policy-blending formalism for shared control (2013, background): “…to struggle with high-cost local minima in the complex spaces induced by manipulation tasks, this issue can be alleviated to a certain extent by learning to place the optimizer in good basins of attraction from prior experience (Jetchev and Toussaint, 2009; Dragan et al., 2011a; Dey et al., 2012).”
Learning preferences for manipulation tasks from online coactive feedback (2015, background): “However, for high DoF manipulators the sampling based approach [6, 12] maintains tractability of the problem, hence we take this approach.”
Multiple Choice Learning: Learning to Produce Multiple Structured Outputs (2012, methodology): “[9] proposed an algorithm that learns a multi-class classifier for each “slot” in a M -Best list, and provide a formal regret reduction from submodular sequence optimization.”