Silver-Bullet-3D at ManiSkill 2021: Learning-from-Demonstrations and Heuristic Rule-based Methods for Object Manipulation

This paper presents an overview and comparative analysis of our systems designed for the following two tracks in SAPIEN ManiSkill Challenge 2021: No Interaction Track: The No Interaction track targets for learning policies from pre-collected demonstration trajectories. We investigate both imitation learning-based approach, i.e., imitating the observed behavior using classical supervised learning techniques, and offline reinforcement learning-based approaches, for this track. Moreover, the geometry and texture structures of objects and robotic arms are exploited via Transformer-based networks to facilitate imitation learning. No Restriction Track: In this track, we design a Heuristic Rule-based Method (HRM) to trigger high-quality object manipulation by decomposing the task into a series of sub-tasks. For each sub-task, the simple rule-based controlling strategies are adopted to predict actions that can be applied to robotic arms. To ease the implementations of our systems, all the source codes and pre-trained models are available at \url{https://github.com/caiqi/Silver-Bullet-3D/}.

Embed to Control Partially Observed Systems: Representation Learning with Provable Sample Efficiency

Reinforcement learning in partially observed Markov decision processes (POMDPs) faces two challenges. (i) It often takes the full history to predict the future, which induces a sample complexity that scales exponentially with the horizon. (ii) The observation and state spaces are often continuous, which induces a sample complexity that scales exponentially with the extrinsic dimension. Addressing such challenges requires learning a minimal but sufficient representation of the observation and state histories by exploiting the structure of the POMDP. To this end, we propose a reinforcement learning algorithm named Embed to Control (ETC), which learns the representation at two levels while optimizing the policy.~(i) For each step, ETC learns to represent the state with a low-dimensional feature, which factorizes the transition kernel. (ii) Across multiple steps, ETC learns to represent the full history with a low-dimensional embedding, which assembles the per-step feature. We integrate (i) and (ii) in a unified framework that allows a variety of estimators (including maximum likelihood estimators and generative adversarial networks). For a class of POMDPs with a low-rank structure in the transition kernel, ETC attains an $O(1/\epsilon^2)$ sample complexity that scales polynomially with the horizon and the intrinsic dimension (that is, the rank). Here $\epsilon$ is the optimality gap. To our best knowledge, ETC is the first sample-efficient algorithm that bridges representation learning and policy optimization in POMDPs with infinite observation and state spaces.

Sample-Efficient Reinforcement Learning for POMDPs with Linear Function Approximations

Despite the success of reinforcement learning (RL) for Markov decision processes (MDPs) with function approximation, most RL algorithms easily fail if the agent only has partial observations of the state. Such a setting is often modeled as a partially observable Markov decision process (POMDP). Existing sample-efficient algorithms for POMDPs are restricted to the tabular setting where the state and observation spaces are finite. In this paper, we make the first attempt at tackling the tension between function approximation and partial observability. In specific, we focus on a class of undercomplete POMDPs with linear function approximations, which allows the state and observation spaces to be infinite. For such POMDPs, we show that the optimal policy and value function can be characterized by a sequence of finite-memory Bellman operators. We propose an RL algorithm that constructs optimistic estimators of these operators via reproducing kernel Hilbert space (RKHS) embedding. Moreover, we theoretically prove that the proposed algorithm finds an $\varepsilon$-optimal policy with $\tilde O (1/\varepsilon^2)$ episodes of exploration. Also, this sample complexity only depends on the intrinsic dimension of the POMDP polynomially and is independent of the size of the state and observation spaces. To our best knowledge, we develop the first provably sample-efficient algorithm for POMDPs with function approximation.

A Low Rank Promoting Prior for Unsupervised Contrastive Learning

Unsupervised learning is just at a tipping point where it could really take off. Among these approaches, contrastive learning has seen tremendous progress and led to state-of-the-art performance. In this paper, we construct a novel probabilistic graphical model that effectively incorporates the low rank promoting prior into the framework of contrastive learning, referred to as LORAC. In contrast to the existing conventional self-supervised approaches that only considers independent learning, our hypothesis explicitly requires that all the samples belonging to the same instance class lie on the same subspace with small dimension. This heuristic poses particular joint learning constraints to reduce the degree of freedom of the problem during the search of the optimal network parameterization. Most importantly, we argue that the low rank prior employed here is not unique, and many different priors can be invoked in a similar probabilistic way, corresponding to different hypotheses about underlying truth behind the contrastive features. Empirical evidences show that the proposed algorithm clearly surpasses the state-of-the-art approaches on multiple benchmarks, including image classification, object detection, instance segmentation and keypoint detection.

A Pose-only Solution to Visual Reconstruction and Navigation

Visual navigation and three-dimensional (3D) scene reconstruction are essential for robotics to interact with the surrounding environment. Large-scale scenes and critical camera motions are great challenges facing the research community to achieve this goal. We raised a pose-only imaging geometry framework and algorithms that can help solve these challenges. The representation is a linear function of camera global translations, which allows for efficient and robust camera motion estimation. As a result, the spatial feature coordinates can be analytically reconstructed and do not require nonlinear optimization. Experiments demonstrate that the computational efficiency of recovering the scene and associated camera poses is significantly improved by 2-4 orders of magnitude. This solution might be promising to unlock real-time 3D visual computing in many forefront applications.

Segmenting Epipolar Line

Identifying feature correspondence between two images is a fundamental procedure in three-dimensional computer vision. Usually the feature search space is confined by the epipolar line. Using the cheirality constraint, this paper finds that the feature search space can be restrained to one of two or three segments of the epipolar line that are defined by the epipole and a so-called virtual infinity point.

Joint Contrastive Learning with Infinite Possibilities

This paper explores useful modifications of the recent development in contrastive learning via novel probabilistic modeling. We derive a particular form of contrastive loss named Joint Contrastive Learning (JCL). JCL implicitly involves the simultaneous learning of an infinite number of query-key pairs, which poses tighter constraints when searching for invariant features. We derive an upper bound on this formulation that allows analytical solutions in an end-to-end training manner. While JCL is practically effective in numerous computer vision applications, we also theoretically unveil the certain mechanisms that govern the behavior of JCL. We demonstrate that the proposed formulation harbors an innate agency that strongly favors similarity within each instance-specific class, and therefore remains advantageous when searching for discriminative features among distinct instances. We evaluate these proposals on multiple benchmarks, demonstrating considerable improvements over existing algorithms. Code is publicly available at: https://github.com/caiqi/Joint-Contrastive-Learning.

A Self Contour-based Rotation and Translation-Invariant Transformation for Point Clouds Recognition

Recently, several direct processing point cloud models have achieved state-of-the-art performances for classification and segmentation tasks. However, these methods lack rotation robustness, and their performances degrade severely under random rotations, failing to extend to real-world applications with varying orientations. To address this problem, we propose a method named Self Contour-based Transformation (SCT), which can be flexibly integrated into a variety of existing point cloud recognition models against arbitrary rotations without any extra modifications. The SCT provides efficient and mathematically proved rotation and translation invariance by introducing Rotation and Translation-Invariant Transformation. It linearly transforms Cartesian coordinates of points to the self contour-based rotation-invariant representations while maintaining the global geometric structure. Moreover, to enhance discriminative feature extraction, the Frame Alignment module is further introduced, aiming to capture contours and transform self contour-based frames to the intra-class frame. Extensive experimental results and mathematical analyses show that the proposed method outperforms the state-of-the-art approaches under arbitrary rotations without any rotation augmentation on standard benchmarks, including ModelNet40, ScanObjectNN and ShapeNet.

On the Global Optimality of Model-Agnostic Meta-Learning

Model-agnostic meta-learning (MAML) formulates meta-learning as a bilevel optimization problem, where the inner level solves each subtask based on a shared prior, while the outer level searches for the optimal shared prior by optimizing its aggregated performance over all the subtasks. Despite its empirical success, MAML remains less understood in theory, especially in terms of its global optimality, due to the nonconvexity of the meta-objective (the outer-level objective). To bridge such a gap between theory and practice, we characterize the optimality gap of the stationary points attained by MAML for both reinforcement learning and supervised learning, where the inner-level and outer-level problems are solved via first-order optimization methods. In particular, our characterization connects the optimality gap of such stationary points with (i) the functional geometry of inner-level objectives and (ii) the representation power of function approximators, including linear models and neural networks. To the best of our knowledge, our analysis establishes the global optimality of MAML with nonconvex meta-objectives for the first time.