Factored Pose Estimation of Articulated Objects using Efficient Nonparametric Belief Propagation

Abstract

Robots working in human environments often encounter a wide range of articulated objects, such as tools, cabinets, and other jointed objects. Such articulated objects can take an infinite number of possible poses, as a point in a potentially high-dimensional continuous space. A robot must perceive this continuous pose to manipulate the object to a desired pose. This problem of perception and manipulation of articulated objects remains a challenge due to its high dimensionality and multi-modal uncertainty. In this paper, we propose a factored approach to estimate the poses of articulated objects using an efficient nonparametric belief propagation algorithm. We consider inputs as geometrical models with articulation constraints, and observed RGBD sensor data. The proposed framework produces object-part pose beliefs iteratively. The problem is formulated as a pairwise Markov Random Field (MRF) where each hidden node (continuous pose variable) is an observed object-part's pose and the edges denote the articulation constraints between the parts. We propose articulated pose estimation by Pull Message Passing algorithm for Nonparametric Belief Propagation (PMPNBP) and evaluate its convergence properties over scenes with articulated objects.

CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds

“…Leveraging Kinematic Chain Constraints For articulated objects, our method focuses on per-part tracking without explicitly leveraging joint constraints at test time, though the constraints are implicitly enforced in training losses. Prior works leverage these constraints in instance-level tracking [18,10] and category-level tracking [14]. [18] and [10] assume perfect knowledge of joint parameters (axis orientation and pivot point positions) and treat them as a hard constraint.…”

“…Prior works leverage these constraints in instance-level tracking [18,10] and category-level tracking [14]. [18] and [10] assume perfect knowledge of joint parameters (axis orientation and pivot point positions) and treat them as a hard constraint. In the category-level setting, however, joint parameters are unknown and difficult to predict due to occlusions, especially for pivot point predictions.…”

A Multi-Step CNN-Based Estimation of Aircraft Landing Gear Angles

“…These methods are constrained by the need to model accurately. In addition, [28][29][30][31] proposed proposed ways that manipulate and interact with the target and then estimate the deformable target attitude. These methods can assess the attitude of unknown targets, but it takes a lot of time to manipulate and interact with the target, and this can only be applied to simple deformable targets.…”

Probabilistic Inference in Planning for Partially Observable Long Horizon Problems

“…However, their approach limits the scope of partial observability to that of object poses. In contrast, through the use of Nonparametric Belief Propagation [5], [6] on the constraint network, our approach provides the avenue for incorporating arbitrary uncertainty models on any of the variables whose value is to be inferred. We evaluate our approach against SS-Replan [12] in our experiments.…”

“…Neither of these approaches explicitly accounts for the uncertainty distributions of the continuous action parameters such as uncertainty in the pose estimates and in the robot's joint configurations. SHY-COBRA instead solves the H-CSP using Pull Message Passing for Nonparametric Belief Propagation [5], [6] because of its natural ability to account for the arbitrary uncertainty models of the continuous variables. The robot executes the actions in turn and replans when necessary.…”