A core challenge for an autonomous agent acting in the real world is to adapt its repertoire of skills to cope with its noisy perception and dynamics. To scale learning of skills to long-horizon tasks, robots should be able to learn and later refine their skills in a structured manner through trajectories rather than making instantaneous decisions individually at each time step. To this end, we propose the Soft Actor-Critic Gaussian Mixture Model (SAC-GMM), a novel hybrid approach that learns robot skills through a dynamical system and adapts the learned skills in their own trajectory distribution space through interactions with the environment. Our approach combines classical robotics techniques of learning from demonstration with the deep reinforcement learning framework and exploits their complementary nature. We show that our method utilizes sensors solely available during the execution of preliminarily learned skills to extract relevant features that lead to faster skill refinement. Extensive evaluations in both simulation and real-world environments demonstrate the effectiveness of our method in refining robot skills by leveraging physical interactions, high-dimensional sensory data, and sparse task completion rewards. Videos, code, and pre-trained models are available at //sac-gmm.cs.uni-freiburg.de.
相關內容
Multi-task learning (MTL) encapsulates multiple learned tasks in a single model and often lets those tasks learn better jointly. However, when deploying MTL onto those real-world systems that are often resource-constrained or latency-sensitive, two prominent challenges arise: (i) during training, simultaneously optimizing all tasks is often difficult due to gradient conflicts across tasks; (ii) at inference, current MTL regimes have to activate nearly the entire model even to just execute a single task. Yet most real systems demand only one or two tasks at each moment, and switch between tasks as needed: therefore such all tasks activated inference is also highly inefficient and non-scalable. In this paper, we present a model-accelerator co-design framework to enable efficient on-device MTL. Our framework, dubbed M$^3$ViT, customizes mixture-of-experts (MoE) layers into a vision transformer (ViT) backbone for MTL, and sparsely activates task-specific experts during training. Then at inference with any task of interest, the same design allows for activating only the task-corresponding sparse expert pathway, instead of the full model. Our new model design is further enhanced by hardware-level innovations, in particular, a novel computation reordering scheme tailored for memory-constrained MTL that achieves zero-overhead switching between tasks and can scale to any number of experts. When executing single-task inference, M$^{3}$ViT achieves higher accuracies than encoder-focused MTL methods, while significantly reducing 88% inference FLOPs. When implemented on a hardware platform of one Xilinx ZCU104 FPGA, our co-design framework reduces the memory requirement by 2.4 times, while achieving energy efficiency up to 9.23 times higher than a comparable FPGA baseline. Code is available at: //github.com/VITA-Group/M3ViT.
While combining imitation learning (IL) and reinforcement learning (RL) is a promising way to address poor sample efficiency in autonomous behavior acquisition, methods that do so typically assume that the requisite behavior demonstrations are provided by an expert that behaves optimally with respect to a task reward. If, however, suboptimal demonstrations are provided, a fundamental challenge appears in that the demonstration-matching objective of IL conflicts with the return-maximization objective of RL. This paper introduces D-Shape, a new method for combining IL and RL that uses ideas from reward shaping and goal-conditioned RL to resolve the above conflict. D-Shape allows learning from suboptimal demonstrations while retaining the ability to find the optimal policy with respect to the task reward. We experimentally validate D-Shape in sparse-reward gridworld domains, showing that it both improves over RL in terms of sample efficiency and converges consistently to the optimal policy in the presence of suboptimal demonstrations.
Prior feature transformation based approaches to Unsupervised Domain Adaptation (UDA) employ the deep features extracted by pre-trained deep models without fine-tuning them on the specific source or target domain data for a particular domain adaptation task. In contrast, end-to-end learning based approaches optimise the pre-trained backbones and the customised adaptation modules simultaneously to learn domain-invariant features for UDA. In this work, we explore the potential of combining fine-tuned features and feature transformation based UDA methods for improved domain adaptation performance. Specifically, we integrate the prevalent progressive pseudo-labelling techniques into the fine-tuning framework to extract fine-tuned features which are subsequently used in a state-of-the-art feature transformation based domain adaptation method SPL (Selective Pseudo-Labeling). Thorough experiments with multiple deep models including ResNet-50/101 and DeiT-small/base are conducted to demonstrate the combination of fine-tuned features and SPL can achieve state-of-the-art performance on several benchmark datasets.
QA models with lifelong learning (LL) abilities are important for practical QA applications, and architecture-based LL methods are reported to be an effective implementation for these models. However, it is non-trivial to extend previous approaches to QA tasks since they either require access to task identities in the testing phase or do not explicitly model samples from unseen tasks. In this paper, we propose Diana: a dynamic architecture-based lifelong QA model that tries to learn a sequence of QA tasks with a prompt enhanced language model. Four types of hierarchically organized prompts are used in Diana to capture QA knowledge from different granularities. Specifically, we dedicate task-level prompts to capture task-specific knowledge to retain high LL performances and maintain instance-level prompts to learn knowledge shared across different input samples to improve the model's generalization performance. Moreover, we dedicate separate prompts to explicitly model unseen tasks and introduce a set of prompt key vectors to facilitate knowledge sharing between tasks. Extensive experiments demonstrate that Diana outperforms state-of-the-art lifelong QA models, especially in handling unseen tasks.
Deep Reinforcement Learning (RL) is mainly studied in a setting where the training and the testing environments are similar. But in many practical applications, these environments may differ. For instance, in control systems, the robot(s) on which a policy is learned might differ from the robot(s) on which a policy will run. It can be caused by different internal factors (e.g., calibration issues, system attrition, defective modules) or also by external changes (e.g., weather conditions). There is a need to develop RL methods that generalize well to variations of the training conditions. In this article, we consider the simplest yet hard to tackle generalization setting where the test environment is unknown at train time, forcing the agent to adapt to the system's new dynamics. This online adaptation process can be computationally expensive (e.g., fine-tuning) and cannot rely on meta-RL techniques since there is just a single train environment. To do so, we propose an approach where we learn a subspace of policies within the parameter space. This subspace contains an infinite number of policies that are trained to solve the training environment while having different parameter values. As a consequence, two policies in that subspace process information differently and exhibit different behaviors when facing variations of the train environment. Our experiments carried out over a large variety of benchmarks compare our approach with baselines, including diversity-based methods. In comparison, our approach is simple to tune, does not need any extra component (e.g., discriminator) and learns policies able to gather a high reward on unseen environments.
Given an untrimmed video and a language query depicting a specific temporal moment in the video, video grounding aims to localize the time interval by understanding the text and video simultaneously. One of the most challenging issues is an extremely time- and cost-consuming annotation collection, including video captions in a natural language form and their corresponding temporal regions. In this paper, we present a simple yet novel training framework for video grounding in the zero-shot setting, which learns a network with only video data without any annotation. Inspired by the recent language-free paradigm, i.e. training without language data, we train the network without compelling the generation of fake (pseudo) text queries into a natural language form. Specifically, we propose a method for learning a video grounding model by selecting a temporal interval as a hypothetical correct answer and considering the visual feature selected by our method in the interval as a language feature, with the help of the well-aligned visual-language space of CLIP. Extensive experiments demonstrate the prominence of our language-free training framework, outperforming the existing zero-shot video grounding method and even several weakly-supervised approaches with large margins on two standard datasets.
Reinforcement learning has been demonstrated as a flexible and effective approach for learning a range of continuous control tasks, such as those used by robots to manipulate objects in their environment. But in robotics particularly, real-world rollouts are costly, and sample efficiency can be a major limiting factor when learning a new skill. In game environments, the use of world models has been shown to improve sample efficiency while still achieving good performance, especially when images or other rich observations are provided. In this project, we explore the use of a world model in a deformable robotic manipulation task, evaluating its effect on sample efficiency when learning to fold a cloth in simulation. We compare the use of RGB image observation with a feature space leveraging built-in structure (keypoints representing the cloth configuration), a common approach in robot skill learning, and compare the impact on task performance and learning efficiency with and without the world model. Our experiments showed that the usage of keypoints increased the performance of the best model on the task by 50%, and in general, the use of a learned or constructed reduced feature space improved task performance and sample efficiency. The use of a state transition predictor(MDN-RNN) in our world models did not have a notable effect on task performance.
Accurate value estimates are important for off-policy reinforcement learning. Algorithms based on temporal difference learning typically are prone to an over- or underestimation bias building up over time. In this paper, we propose a general method called Adaptively Calibrated Critics (ACC) that uses the most recent high variance but unbiased on-policy rollouts to alleviate the bias of the low variance temporal difference targets. We apply ACC to Truncated Quantile Critics, which is an algorithm for continuous control that allows regulation of the bias with a hyperparameter tuned per environment. The resulting algorithm adaptively adjusts the parameter during training rendering hyperparameter search unnecessary and sets a new state of the art on the OpenAI gym continuous control benchmark among all algorithms that do not tune hyperparameters for each environment. ACC further achieves improved results on different tasks from the Meta-World robot benchmark. Additionally, we demonstrate the generality of ACC by applying it to TD3 and showing an improved performance also in this setting.
Motion planning and control are crucial components of robotics applications. Here, spatio-temporal hard constraints like system dynamics and safety boundaries (e.g., obstacles in automated driving) restrict the robot's motions. Direct methods from optimal control solve a constrained optimization problem. However, in many applications finding a proper cost function is inherently difficult because of the weighting of partially conflicting objectives. On the other hand, Imitation Learning (IL) methods such as Behavior Cloning (BC) provide a intuitive framework for learning decision-making from offline demonstrations and constitute a promising avenue for planning and control in complex robot applications. Prior work primarily relied on soft-constraint approaches, which use additional auxiliary loss terms describing the constraints. However, catastrophic safety-critical failures might occur in out-of-distribution (OOD) scenarios. This work integrates the flexibility of IL with hard constraint handling in optimal control. Our approach constitutes a general framework for constraint robotic motion planning and control using offline IL. Hard constraints are integrated into the learning problem in a differentiable manner, via explicit completion and gradient-based correction. Simulated experiments of mobile robot navigation and automated driving provide evidence for the performance of the proposed method.
Designing reinforcement learning (RL) agents is typically a difficult process that requires numerous design iterations. Learning can fail for a multitude of reasons, and standard RL methods provide too few tools to provide insight into the exact cause. In this paper, we show how to integrate value decomposition into a broad class of actor-critic algorithms and use it to assist in the iterative agent-design process. Value decomposition separates a reward function into distinct components and learns value estimates for each. These value estimates provide insight into an agent's learning and decision-making process and enable new training methods to mitigate common problems. As a demonstration, we introduce SAC-D, a variant of soft actor-critic (SAC) adapted for value decomposition. SAC-D maintains similar performance to SAC, while learning a larger set of value predictions. We also introduce decomposition-based tools that exploit this information, including a new reward influence metric, which measures each reward component's effect on agent decision-making. Using these tools, we provide several demonstrations of decomposition's use in identifying and addressing problems in the design of both environments and agents. Value decomposition is broadly applicable and easy to incorporate into existing algorithms and workflows, making it a powerful tool in an RL practitioner's toolbox.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191