The International Journal of Robotics Research

We consider a semi-supervised approach to the problem of track classification in dense three-dimensional range data. This problem involves the classification of objects that have been segmented and tracked without the use of a class-specific tracker. This paper is an extended version of our previous work.

We propose a method based on the expectation–maximization algorithm: iteratively (1) train a classifier, and (2) extract useful training examples from unlabeled data by exploiting tracking information. We evaluate our method on a large multiclass problem in dense range data collected from natural street scenes.

When given only three hand-labeled training tracks of each object class, the final accuracy of the semi-supervised algorithm is comparable to that of the fully supervised equivalent which uses two orders of magnitude more. Further, we show experimentally that the accuracy of a classifier considered as a function of human labeling effort can be substantially improved using this method. Finally, we show that a simple algorithmic speedup based on incrementally updating a boosting classifier can reduce learning time by a factor of three.

We model the full dynamics of a rigid part in three-point frictional sliding contact with a flat rigid six-degree-of-freedom (6-DoF) plate. When the plate moves periodically, we show the part’s dynamics are well approximated by a first-order system represented by an asymptotic velocity field that maps part configurations in SE(2) to unique velocities (linear and angular) in ℝ² . The form of the asymptotic velocity field depends on the plate’s motion, the location and friction coefficient of each contact point, and the inertial properties of the part. Asymptotic velocity vectors in the field approximate the part’s cycle-averaged velocity at each configuration and are independent of time or the system’s initial state. For the special case of a rigid part with infinitesimal thickness, we prove that asymptotic velocities are always unique and well defined. With the ability to program arbitrary periodic plate motions, part manipulation reduces to finding plate motions that generate asymptotic velocity fields to accomplish desired tasks. Several fields useful for manipulation tasks (e.g. sensorless part alignment) are verified in simulation and experiment.

In this paper we develop methods for maximizing the throughput of a mobility-on-demand urban transportation system. We consider a finite group of shared vehicles, located at a set of stations. Users arrive at the stations, pickup vehicles, and drive (or are driven) to their destination station where they drop-off the vehicle. When some origins and destinations are more popular than others, the system will inevitably become out of balance: vehicles will build up at some stations, and become depleted at others. We propose a robotic solution to this rebalancing problem that involves empty robotic vehicles autonomously driving between stations. Specifically, we utilize a fluid model for the customers and vehicles in the system. Then, we develop a rebalancing policy that lets every station reach an equilibrium in which there are excess vehicles and no waiting customers and that minimizes the number of robotic vehicles performing rebalancing trips. We show that the optimal rebalancing policy can be found as the solution to a linear program. We use this solution to develop a real-time rebalancing policy which can operate in highly variable environments. Finally, we verify policy performance in a simulated mobility-on-demand environment and in hardware experiments.

This paper is concerned with motion planning for non-linear robotic systems operating in constrained environments. A method for computing high-quality trajectories is proposed building upon recent developments in sampling-based motion planning and stochastic optimization. The idea is to equip sampling-based methods with a probabilistic model that serves as a sampling distribution and to incrementally update the model during planning using data collected by the algorithm. At the core of the approach lies the cross-entropy method for the estimation of rare-event probabilities. The cross-entropy method is combined with recent optimal motion planning methods such as the rapidly exploring random trees (RRT*) in order to handle complex environments. The main goal is to provide a framework for consistent adaptive sampling that correlates the spatial structure of trajectories and their computed costs in order to improve the performance of existing planning methods.

This paper provides an analytical solution to the motion-planning problem for the Snakeboard. Given a desired planar trajectory in the fiber space, an explicit solution is computed for the controllable inputs in the base space that locomote the Snakeboard along a given trajectory. The motion-planning problem or gait generation problem is solved for the generalized Snakeboard where the orientation of the wheels is not coupled, as well as for several special configurations of the Snakeboard.

This paper digs into the relationship between cages and grasps of a rigid body. In particular, it considers the use of cages as waypoints to grasp an object. We introduce the concept of pregrasping cages, caging configurations from which an object can be grasped without first breaking the cage. For two-fingered manipulators, all cages are pregrasping cages and, consequently, useful waypoints to grasp an object. A contribution of this paper is to show that the same does not hold for more than two fingers. A second contribution is to show how to overcome that limitation. We explore the natural generalization of the well-understood squeezing/stretching characterization of two-finger cages to arbitrary workspace dimension, arbitrary object shapes without holes, and arbitrary number of point fingers, and exploit it to give sufficient conditions for a cage to be a pregrasping cage. As a product of that generalization, we introduce grasping functions: scalar functions defined on the finger formation that control the process of going from a cage to a grasp. We finish the paper by establishing an analogy between the role of grasping functions in grasping and that of Lyapunov functions in stability theory.

Recent trends pushing robots into unstructured environments with limited sensors have motivated considerable work on planning under uncertainty and stochastic optimal control, but these methods typically do not provide guaranteed performance. Here we consider the problem of bounding the probability of failure (defined as leaving a finite region of state space) over a finite time for stochastic non-linear systems with continuous state. Our approach searches for exponential barrier functions that provide bounds using a variant of the classical supermartingale result. We provide a relaxation of this search to a semidefinite program, yielding an efficient algorithm that provides rigorous upper bounds on the probability of failure for the original non-linear system. We give a number of numerical examples in both discrete and continuous time that demonstrate the effectiveness of the approach.