Anna Petrovskaya

Model based vehicle detection and tracking for autonomous urban driving

Situational awareness is crucial for autonomous driving in urban environments. This paper describes the moving vehicle detection and tracking module that we developed for our autonomous driving robot Junior. The robot won second place in the Urban Grand Challenge, an autonomous driving race organized by the U.S. Government in 2007. The module provides reliable detection and tracking of moving vehicles from a high-speed moving platform using laser range finders. Our approach models both dynamic and geometric properties of the tracked vehicles and estimates them using a single Bayes filter per vehicle. We present the notion of motion evidence, which allows us to overcome the low signal-to-noise ratio that arises during rapid detection of moving vehicles in noisy urban environments. Furthermore, we show how to build consistent and efficient 2D representations out of 3D range data and how to detect poorly visible black vehicles. Experimental validation includes the most challenging conditions presented at the Urban Grand Challenge as well as other urban settings.

Global Localization of Objects via Touch

Humans are capable of manipulating objects based solely on the sense of touch. For robots to achieve the same feat in unstructured environments, global localization of objects via touch is required. Bayesian approaches provide the means to cope with uncertainties of the real world, but the estimation of the Bayesian posterior for the full six degrees of freedom (6-DOF) global localization problem is computationally prohibitive. We propose an efficient Bayesian approach termed Scaling Series. It is capable of solving the full problem reliably in real time. This is a Monte Carlo approach that performs a series of successive refinements coupled with annealing. We also propose an analytical measurement model, which can be computed efficiently at run time for any object represented as a polygonal mesh. Extensive empirical evaluation shows that Scaling Series drastically outperforms prior approaches. We demonstrate general applicability of the approach on five common solid objects, which are rigidly fixed during the experiments. We also consider 6-DOF localization and tracking of free-standing objects that can move during tactile exploration.

Bayesian estimation for autonomous object manipulation based on tactile sensors

We consider the problem of autonomously estimating position and orientation of an object from tactile data. When initial uncertainty is high, estimation of all six parameters precisely is computationally expensive. We propose an efficient Bayesian approach that is able to estimate all six parameters in both unimodal and multimodal scenarios. The approach is termed scaling series sampling as it estimates the solution region by samples. It performs the search using a series of successive refinements, gradually scaling the precision from low to high. Our approach can be applied to a wide range of manipulation tasks. We demonstrate its portability on two applications: (1) manipulating a box and (2) grasping a door handle

Model Based Vehicle Tracking for Autonomous Driving in Urban Environments

Situational awareness is crucial for autonomous driving in urban environments. This paper describes moving vehicle tracking module that we developed for our autonomous driving robot Junior. The robot won second place in the Urban Grand Challenge, an autonomous driving race organized by the U.S. Government in 2007. The tracking module provides reliable tracking of moving vehicles from a high-speed moving platform using laser range finders. Our approach models both dynamic and geometric properties of the tracked vehicles and estimates them using a single Bayes filter per vehicle. We also show how to build efficient 2D representations out of 3D range data and how to detect poorly visible black vehicles. Experimental validation includes the most challenging conditions presented at the UGC as well as other urban settings.

Probabilistic Estimation of Whole Body Contacts for Multi-Contact Robot Control

Today most robots interact with the surroundings only with their end-effectors. However there are many benefits to utilizing contact along the entire length of robot body and links especially for human-like robots. Existing control strategies for link contact require knowledge of the contact point. In an uncertain environment, locating link contact point is difficult for most robots as they do not possess skin capable of sensing. We propose a probabilistic approach to link contact estimation based on geometric considerations and compliant motions. Since for many robots, link geometry is also uncertain, we broaden our approach to simultaneously estimate link shape and environment contact. Our experimental results demonstrate that efficiency of control is significantly improved by link contact estimation

Identifying physical properties of deformable objects by using particle filters

This paper presents a new approach for estimating physical properties of deformable models from experimental measurements. In contrast to most previous work, we introduce a new method based on particle filters which identifies the different stiffness properties for spring-based models. This approach addresses some important limitations encountered with gradient descent techniques which often converge towards ill solutions or remain fixed in local minima conditions.

Efficient Techniques for Dynamic Vehicle Detection

Fast detection of moving vehicles is crucial for safe autonomous urban driving. We present the vehicle detection algorithm developed for our entry in the Urban Grand Challenge, an autonomous driving race organized by the U.S. Government in 2007. The algorithm provides reliable detection of moving vehicles from a high-speed moving platform using laser range finders. We present the notion of motion evidence, which allows us to overcome the low signal-to-noise ratio that arises during rapid detection of moving vehicles in noisy urban environments. We also present and evaluate an array of optimization techniques that enable accurate detection in real time. Experimental results show empirical validation on data from the most challenging situations presented at the Urban Grand Challenge as well as other urban settings.

Active Manipulation for Perception

This chapter covers perceptual methods in which manipulation is an integral part of perception. These methods face special challenges due to data sparsity and high costs of sensing actions. However, they can also succeed where other perceptual methods fail, for example, in poor-visibility conditions or for learning the physical properties of a scene.