Transfer reinforcement learning aims to improve the sample efficiency of solving unseen new tasks by leveraging experiences obtained from previous tasks. We consider the setting where all tasks (MDPs) share the same environment dynamic except reward function. In this setting, the MDP dynamic is a good knowledge to transfer, which can be inferred by uniformly random policy. However, trajectories generated by uniform random policy are not useful for policy improvement, which impairs the sample efficiency severely. Instead, we observe that the binary MDP dynamic can be inferred from trajectories of any policy which avoids the need of uniform random policy. As the binary MDP dynamic contains the state structure shared over all tasks we believe it is suitable to transfer. Built on this observation, we introduce a method to infer the binary MDP dynamic on-line and at the same time utilize it to guide state embedding learning, which is then transferred to new tasks. We keep state embedding learning and policy learning separately. As a result, the learned state embedding is task and policy agnostic which makes it ideal for transfer learning. In addition, to facilitate the exploration over the state space, we propose a novel intrinsic reward based on the inferred binary MDP dynamic. Our method can be used out-of-box in combination with model-free RL algorithms. We show two instances on the basis of \algo{DQN} and \algo{A2C}. Empirical results of intensive experiments show the advantage of our proposed method in various transfer learning tasks.
Underwater robots in shallow waters usually suffer from strong wave forces, which may frequently exceed robot's control constraints. Learning-based controllers are suitable for disturbance rejection control, but the excessive disturbances heavily affect the state transition in Markov Decision Process (MDP) or Partially Observable Markov Decision Process (POMDP). Also, pure learning procedures on targeted system may encounter damaging exploratory actions or unpredictable system variations, and training exclusively on a prior model usually cannot address model mismatch from the targeted system. In this paper, we propose a transfer learning framework that adapts a control policy for excessive disturbance rejection of an underwater robot under dynamics model mismatch. A modular network of learning policies is applied, composed of a Generalized Control Policy (GCP) and an Online Disturbance Identification Model (ODI). GCP is first trained over a wide array of disturbance waveforms. ODI then learns to use past states and actions of the system to predict the disturbance waveforms which are provided as input to GCP (along with the system state). A transfer reinforcement learning algorithm using Transition Mismatch Compensation (TMC) is developed based on the modular architecture, that learns an additional compensatory policy through minimizing mismatch of transitions predicted by the two dynamics models of the source and target tasks. We demonstrated on a pose regulation task in simulation that TMC is able to successfully reject the disturbances and stabilize the robot under an empirical model of the robot system, meanwhile improve sample efficiency.
This article proposes a transfer reinforcement learning (RL) based adaptive energy managing approach for a hybrid electric vehicle (HEV) with parallel topology. This approach is bi-level. The up-level characterizes how to transform the Q-value tables in the RL framework via driving cycle transformation (DCT). Especially, transition probability matrices (TPMs) of power request are computed for different cycles, and induced matrix norm (IMN) is employed as a critical criterion to identify the transformation differences and to determine the alteration of the control strategy. The lower-level determines how to set the corresponding control strategies with the transformed Q-value tables and TPMs by using model-free reinforcement learning (RL) algorithm. Numerical tests illustrate that the transferred performance can be tuned by IMN value and the transfer RL controller could receive a higher fuel economy. The comparison demonstrates that the proposed strategy exceeds the conventional RL approach in both calculation speed and control performance.
In this paper, we study a transfer reinforcement learning problem where the state transitions and rewards are affected by the environmental context. Specifically, we consider a demonstrator agent that has access to a context-aware policy and can generate transition and reward data based on that policy. These data constitute the experience of the demonstrator. Then, the goal is to transfer this experience, excluding the underlying contextual information, to a learner agent that does not have access to the environmental context, so that they can learn a control policy using fewer samples. It is well known that, disregarding the causal effect of the contextual information, can introduce bias in the transition and reward models estimated by the learner, resulting in a learned suboptimal policy. To address this challenge, in this paper, we develop a method to obtain causal bounds on the transition and reward functions using the demonstrator's data, which we then use to obtain causal bounds on the value functions. Using these value function bounds, we propose new Q learning and UCB-Q learning algorithms that converge to the true value function without bias. We provide numerical experiments for robot motion planning problems that validate the proposed value function bounds and demonstrate that the proposed algorithms can effectively make use of the data from the demonstrator to accelerate the learning process of the learner.
Transfer in Reinforcement Learning (RL) refers to the idea of applying knowledge gained from previous tasks to solving related tasks. Learning a universal value function (Schaul et al., 2015), which generalizes over goals and states, has previously been shown to be useful for transfer. However, successor features are believed to be more suitable than values for transfer (Dayan, 1993; Barreto et al.,2017), even though they cannot directly generalize to new goals. In this paper, we propose (1) Universal Successor Features (USFs) to capture the underlying dynamics of the environment while allowing generalization to unseen goals and (2) a flexible end-to-end model of USFs that can be trained by interacting with the environment. We show that learning USFs is compatible with any RL algorithm that learns state values using a temporal difference method. Our experiments in a simple gridworld and with two MuJoCo environments show that USFs can greatly accelerate training when learning multiple tasks and can effectively transfer knowledge to new tasks.
One of the key approaches to save samples in reinforcement learning (RL) is to use knowledge from an approximate model such as its simulator. However, how much does an approximate model help to learn a near-optimal policy of the true unknown model? Despite numerous empirical studies of transfer reinforcement learning, an answer to this question is still elusive. In this paper, we study the sample complexity of RL while an approximate model of the environment is provided. For an unknown Markov decision process (MDP), we show that the approximate model can effectively reduce the complexity by eliminating sub-optimal actions from the policy searching space. In particular, we provide an algorithm that uses $\widetilde{O}(N/(1-\gamma)^3/\varepsilon^2)$ samples in a generative model to learn an $\varepsilon$-optimal policy, where $\gamma$ is the discount factor and $N$ is the number of near-optimal actions in the approximate model. This can be much smaller than the learning-from-scratch complexity $\widetilde{\Theta}(SA/(1-\gamma)^3/\varepsilon^2)$, where $S$ and $A$ are the sizes of state and action spaces respectively. We also provide a lower bound showing that the above upper bound is nearly-tight if the value gap between near-optimal actions and sub-optimal actions in the approximate model is sufficiently large. Our results provide a very precise characterization of how an approximate model helps reinforcement learning when no additional assumption on the model is posed.
Reinforcement learning (RL) is widely used in autonomous driving tasks and training RL models typically involves in a multi-step process: pre-training RL models on simulators, uploading the pre-trained model to real-life robots, and fine-tuning the weight parameters on robot vehicles. This sequential process is extremely time-consuming and more importantly, knowledge from the fine-tuned model stays local and can not be re-used or leveraged collaboratively. To tackle this problem, we present an online federated RL transfer process for real-time knowledge extraction where all the participant agents make corresponding actions with the knowledge learned by others, even when they are acting in very different environments. To validate the effectiveness of the proposed approach, we constructed a real-life collision avoidance system with Microsoft Airsim simulator and NVIDIA JetsonTX2 car agents, which cooperatively learn from scratch to avoid collisions in indoor environment with obstacle objects. We demonstrate that with the proposed framework, the simulator car agents can transfer knowledge to the RC cars in real-time, with 27% increase in the average distance with obstacles and 42% decrease in the collision counts.
Transfer reinforcement learning (RL) aims at improving the learning efficiency of an agent by exploiting knowledge from other source agents trained on relevant tasks. However, it remains challenging to transfer knowledge between different environmental dynamics without having access to the source environments. In this work, we explore a new challenge in transfer RL, where only a set of source policies collected under diverse unknown dynamics is available for learning a target task efficiently. To address this problem, the proposed approach, MULTI-source POLicy AggRegation (MULTIPOLAR), comprises two key techniques. We learn to aggregate the actions provided by the source policies adaptively to maximize the target task performance. Meanwhile, we learn an auxiliary network that predicts residuals around the aggregated actions, which ensures the target policy's expressiveness even when some of the source policies perform poorly. We demonstrated the effectiveness of MULTIPOLAR through an extensive experimental evaluation across six simulated environments ranging from classic control problems to challenging robotics simulations, under both continuous and discrete action spaces. The demo videos and code are available on the project webpage: https://omron-sinicx.github.io/multipolar/.
Agents trained with deep reinforcement learning algorithms are capable of performing highly complex tasks including locomotion in continuous environments. We investigate transferring the learning acquired in one task to a set of previously unseen tasks. Generalization and overfitting in deep reinforcement learning are not commonly addressed in current transfer learning research. Conducting a comparative analysis without an intermediate regularization step results in underperforming benchmarks and inaccurate algorithm comparisons due to rudimentary assessments. In this study, we propose regularization techniques in deep reinforcement learning for continuous control through the application of sample elimination, early stopping and maximum entropy regularized adversarial learning. First, the importance of the inclusion of training iteration number to the hyperparameters in deep transfer reinforcement learning will be discussed. Because source task performance is not indicative of the generalization capacity of the algorithm, we start by acknowledging the training iteration number as a hyperparameter. In line with this, we introduce an additional step of resorting to earlier snapshots of policy parameters to prevent overfitting to the source task. Then, to generate robust policies, we discard the samples that lead to overfitting via a method we call strict clipping. Furthermore, we increase the generalization capacity in widely used transfer learning benchmarks by using maximum entropy regularization, different critic methods, and curriculum learning in an adversarial setup. Subsequently, we propose maximum entropy adversarial reinforcement learning to increase the domain randomization. Finally, we evaluate the robustness of these methods on simulated robots in target environments where the morphology of the robot, gravity, and tangential friction coefficient of the environment are altered.
In this paper, we show how novel transfer reinforcement learning techniques can be applied to the complex task of target driven navigation using the photorealistic AI2THOR simulator. Specifically, we build on the concept of Universal Successor Features with an A3C agent. We introduce the novel architectural contribution of a Successor Feature Dependant Policy (SFDP) and adopt the concept of Variational Information Bottlenecks to achieve state of the art performance. VUSFA, our final architecture, is a straightforward approach that can be implemented using our open source repository. Our approach is generalizable, showed greater stability in training, and outperformed recent approaches in terms of transfer learning ability.
This paper presents an upgraded, real world application oriented version of gym-gazebo, the Robot Operating System (ROS) and Gazebo based Reinforcement Learning (RL) toolkit, which complies with OpenAI Gym. The content discusses the new ROS 2 based software architecture and summarizes the results obtained using Proximal Policy Optimization (PPO). Ultimately, the output of this work presents a benchmarking system for robotics that allows different techniques and algorithms to be compared using the same virtual conditions. We have evaluated environments with different levels of complexity of the Modular Articulated Robotic Arm (MARA), reaching accuracies in the millimeter scale. The converged results show the feasibility and usefulness of the gym-gazebo 2 toolkit, its potential and applicability in industrial use cases, using modular robots.
Deep reinforcement learning could be used to learn dexterous robotic policies but it is challenging to transfer them to new robots with vastly different hardware properties. It is also prohibitively expensive to learn a new policy from scratch for each robot hardware due to the high sample complexity of modern state-of-the-art algorithms. We propose a novel approach called \textit{Hardware Conditioned Policies} where we train a universal policy conditioned on a vector representation of robot hardware. We considered robots in simulation with varied dynamics, kinematic structure, kinematic lengths and degrees-of-freedom. First, we use the kinematic structure directly as the hardware encoding and show great zero-shot transfer to completely novel robots not seen during training. For robots with lower zero-shot success rate, we also demonstrate that fine-tuning the policy network is significantly more sample-efficient than training a model from scratch. In tasks where knowing the agent dynamics is important for success, we learn an embedding for robot hardware and show that policies conditioned on the encoding of hardware tend to generalize and transfer well. The code and videos are available on the project webpage: https://sites.google.com/view/robot-transfer-hcp.
Deep neural networks are data hungry models and thus face difficulties when attempting to train on small text datasets. Transfer learning is a potential solution but their effectiveness in the text domain is not as explored as in areas such as image analysis. In this paper, we study the problem of transfer learning for text summarization and discuss why existing state-of-the-art models fail to generalize well on other (unseen) datasets. We propose a reinforcement learning framework based on a self-critic policy gradient approach which achieves good generalization and state-of-the-art results on a variety of datasets. Through an extensive set of experiments, we also show the ability of our proposed framework to fine-tune the text summarization model using only a few training samples. To the best of our knowledge, this is the first work that studies transfer learning in text summarization and provides a generic solution that works well on unseen data.
We propose a probabilistic framework to directly insert prior knowledge in reinforcement learning (RL) algorithms by defining the behaviour policy as a Bayesian posterior distribution. Such a posterior combines task specific information with prior knowledge, thus allowing to achieve transfer learning across tasks. The resulting method is flexible and it can be easily incorporated to any standard off-policy and on-policy algorithms, such as those based on temporal differences and policy gradients. We develop a specific instance of this Bayesian transfer RL framework by expressing prior knowledge as general deterministic rules that can be useful in a large variety of tasks, such as navigation tasks. Also, we elaborate more on recent probabilistic and entropy-regularised RL by developing a novel temporal learning algorithm and show how to combine it with Bayesian transfer RL. Finally, we demonstrate our method for solving mazes and show that significant speed ups can be obtained.
Reinforcement Learning methods are capable of solving complex problems, but resulting policies might perform poorly in environments that are even slightly different. In robotics especially, training and deployment conditions often vary and data collection is expensive, making retraining undesirable. Simulation training allows for feasible training times, but on the other hand suffers from a reality-gap when applied in real-world settings. This raises the need of efficient adaptation of policies acting in new environments. We consider this as a problem of transferring knowledge within a family of similar Markov decision processes. For this purpose we assume that Q-functions are generated by some low-dimensional latent variable. Given such a Q-function, we can find a master policy that can adapt given different values of this latent variable. Our method learns both the generative mapping and an approximate posterior of the latent variables, enabling identification of policies for new tasks by searching only in the latent space, rather than the space of all policies. The low-dimensional space, and master policy found by our method enables policies to quickly adapt to new environments. We demonstrate the method on both a pendulum swing-up task in simulation, and for simulation-to-real transfer on a pushing task.
The ability to act in multiple environments and transfer previous knowledge to new situations can be considered a critical aspect of any intelligent agent. Towards this goal, we define a novel method of multitask and transfer learning that enables an autonomous agent to learn how to behave in multiple tasks simultaneously, and then generalize its knowledge to new domains. This method, termed "Actor-Mimic", exploits the use of deep reinforcement learning and model compression techniques to train a single policy network that learns how to act in a set of distinct tasks by using the guidance of several expert teachers. We then show that the representations learnt by the deep policy network are capable of generalizing to new tasks with no prior expert guidance, speeding up learning in novel environments. Although our method can in general be applied to a wide range of problems, we use Atari games as a testing environment to demonstrate these methods.
It is difficult to train a personalized task-oriented dialogue system because the data collected from each individual is often insufficient. Personalized dialogue systems trained on a small dataset can overfit and make it difficult to adapt to different user needs. One way to solve this problem is to consider a collection of multiple users' data as a source domain and an individual user's data as a target domain, and to perform a transfer learning from the source to the target domain. By following this idea, we propose "PETAL"(PErsonalized Task-oriented diALogue), a transfer-learning framework based on POMDP to learn a personalized dialogue system. The system first learns common dialogue knowledge from the source domain and then adapts this knowledge to the target user. This framework can avoid the negative transfer problem by considering differences between source and target users. The policy in the personalized POMDP can learn to choose different actions appropriately for different users. Experimental results on a real-world coffee-shopping data and simulation data show that our personalized dialogue system can choose different optimal actions for different users, and thus effectively improve the dialogue quality under the personalized setting.
We present Shapechanger, a library for transfer reinforcement learning specifically designed for robotic tasks. We consider three types of knowledge transfer---from simulation to simulation, from simulation to real, and from real to real---and a wide range of tasks with continuous states and actions. Shapechanger is under active development and open-sourced at: https://github.com/seba-1511/shapechanger/.
A key challenge in complex visuomotor control is learning abstract representations that are effective for specifying goals, planning, and generalization. To this end, we introduce universal planning networks (UPN). UPNs embed differentiable planning within a goal-directed policy. This planning computation unrolls a forward model in a latent space and infers an optimal action plan through gradient descent trajectory optimization. The plan-by-gradient-descent process and its underlying representations are learned end-to-end to directly optimize a supervised imitation learning objective. We find that the representations learned are not only effective for goal-directed visual imitation via gradient-based trajectory optimization, but can also provide a metric for specifying goals using images. The learned representations can be leveraged to specify distance-based rewards to reach new target states for model-free reinforcement learning, resulting in substantially more effective learning when solving new tasks described via image-based goals. We were able to achieve successful transfer of visuomotor planning strategies across robots with significantly different morphologies and actuation capabilities.
The objective of transfer reinforcement learning is to generalize from a set of previous tasks to unseen new tasks. In this work, we focus on the transfer scenario where the dynamics among tasks are the same, but their goals differ. Although general value function (Sutton et al., 2011) has been shown to be useful for knowledge transfer, learning a universal value function can be challenging in practice. To attack this, we propose (1) to use universal successor representations (USR) to represent the transferable knowledge and (2) a USR approximator (USRA) that can be trained by interacting with the environment. Our experiments show that USR can be effectively applied to new tasks, and the agent initialized by the trained USRA can achieve the goal considerably faster than random initialization.