Interactive non-prehensile manipulation for grasping via POMDPs

2017 IEEE International Conference on Robotics and Automation (ICRA)

Klingensmith

Srinivasa

et al. 2017

Abstract-We estimate the state a noisy robot arm and underactuated hand using an Implicit Manifold Particle Filter (MPF) informed by touch sensors. As the robot touches the world, its state space collapses to a contact manifold that we represent implicitly using a signed distance field. This allows us to extend the MPF to higher (six or more) dimensional state spaces. Earlier work (which explicitly represents the contact manifold) only shows the MPF in two or three dimensions [1]. Through a series of experiments, we show that the implicit MPF converges faster and is more accurate than a conventional particle filter during periods of persistent contact. We present three methods of sampling the implicit contact manifold, and compare them in experiments.

Section: Related Workmentioning

confidence: 99%

The manifold particle filter for state estimation on high-dimensional implicit manifolds

2017 IEEE International Conference on Robotics and Automation (ICRA)

Klingensmith

Srinivasa

et al. 2017

“…POMDP solvers reason about uncertainty and can integrate observations, but do not easily generalize to continuous action spaces or non-additive reward functions. Most applications of POMDPs to manipulation rely on discretization [17,18,26] or receding horizon planning [34,41].…”

Section: Related Workmentioning

confidence: 99%

Robust trajectory selection for rearrangement planning as a multi-armed bandit problem

2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

King

Pollard

et al. 2015

Abstract-We present an algorithm for generating openloop trajectories that solve the problem of rearrangement planning under uncertainty. We frame this as a selection problem where the goal is to choose the most robust trajectory from a finite set of candidates. We generate each candidate using a kinodynamic state space planner and evaluate it using noisy rollouts.Our key insight is we can formalize the selection problem as the "best arm" variant of the multi-armed bandit problem. We use the successive rejects algorithm to efficiently allocate rollouts between candidate trajectories given a rollout budget. We show that the successive rejects algorithm identifies the best candidate using fewer rollouts than a baseline algorithm in simulation. We also show that selecting a good candidate increases the likelihood of successful execution on a real robot. I. IntroductionWe explore the rearrangement planning problem [11] where a robot must rearrange several objects in a cluttered environment to achieve a goal ( Fig.1-Top). Recent work has used simple physics models, like quasistatic pushing [8,9,22,32,39,43], to quickly produce efficient and intricate plans. However, the intricacy of these plans makes them particularly sensitive to uncertainty in object pose, physics parameters, and trajectory execution.We address this problem by generating open-loop trajectories that are robust to uncertainty. However, this is particularly hard for rearrangement planning.First, the problem is set in a high-dimensional space with continuous actions. Second, contact causes physics to evolve in complex, non-linear ways and quickly leads to multi-modal and non-smooth distributions [25,26,34]. Third, finding good trajectories is inherently hard: most trajectories achieve success with zero probability.As a consequence, this problem lies outside the domains of standard conformant planning [19,37], POMDP [20] and policy search [7] algorithms.In response to these challenges, we propose a domain-agnostic algorithm that only requires: (1) a stochastic method of generating trajectories, (2) the ability to forward-simulate the system's dynamics, and (3) the capability of testing whether an execution is successful.Exploiting the fact that we can quickly generate feasible state space trajectories [22], we formulate rearrangement planning under uncertainty (sections III and IV)

“…It may be possible to leverage this knowledge in a special-purpose POMDP solver. We are encouraged by recent work-including our own (Koval et al 2014)-that has achieved promising results in grasping (Hsiao 2009, Platt et al 2011 and non-prehensile manipulation (Horowitz & Burdick 2013) using a POMDP formulation of the problem.…”

Section: Real-time Feedbackmentioning

confidence: 99%

Pose estimation for planar contact manipulation with manifold particle filters

The International Journal of Robotics Research

Pollard

Srinivasa

2015

We investigate the problem of using contact sensors to estimate the pose of an object during planar pushing by a fixed-shape hand. Contact sensors are unique because they inherently discriminate between "contact" and "no-contact" configurations. As a result, the set of object configurations that activates a sensor constitutes a lower-dimensional contact manifold in the configuration space of the object. This causes conventional state estimation methods, such as the particle filter, to perform poorly during periods of contact due to particle starvation.In this paper, we introduce the manifold particle filter as a principled way of solving the state estimation problem when the state moves between multiple manifolds of different dimensionality. The manifold particle filter avoids particle starvation during contact by adaptively sampling particles that reside on the contact manifold from the dual proposal distribution. We describe three techniques-one analytical, and two sample-based-of sampling from the dual proposal distribution and compare their relative strengths and weaknesses. We present simulation results that show that all three techniques outperform the conventional particle filter in both speed and accuracy. Additionally, we implement the manifold particle filter on a real robot and show that it successfully tracks the pose of a pushed object using commercially available tactile sensors.