It is crucial to address the following issues for ubiquitous robotics manipulation applications: (a) vision-based manipulation tasks require the robot to visually learn and understand the object with rich information like dense object descriptors; and (b) sim-to-real transfer in robotics aims to close the gap between simulated and real data. In this paper, we present Sim-to-Real Dense Object Nets (SRDONs), a dense object descriptor that not only understands the object via appropriate representation but also maps simulated and real data to a unified feature space with pixel consistency. We proposed an object-to-object matching method for image pairs from different scenes and different domains. This method helps reduce the effort of training data from real-world by taking advantage of public datasets, such as GraspNet. With sim-to-real object representation consistency, our SRDONs can serve as a building block for a variety of sim-to-real manipulation tasks. We demonstrate in experiments that pre-trained SRDONs significantly improve performances on unseen objects and unseen visual environments for various robotic tasks with zero real-world training.

Deep Reinforcement Learning has been successfully applied to learn robotic control. However, the corresponding algorithms struggle when applied to problems where the agent is only rewarded after achieving a complex task. In this context, using demonstrations can significantly speed up the learning process, but demonstrations can be costly to acquire. In this paper, we propose to leverage a sequential bias to learn control policies for complex robotic tasks using a single demonstration. To do so, our method learns a goal-conditioned policy to control a system between successive low-dimensional goals. This sequential goal-reaching approach raises a problem of compatibility between successive goals: we need to ensure that the state resulting from reaching a goal is compatible with the achievement of the following goals. To tackle this problem, we present a new algorithm called DCIL-II. We show that DCIL-II can solve with unprecedented sample efficiency some challenging simulated tasks such as humanoid locomotion and stand-up as well as fast running with a simulated Cassie robot. Our method leveraging sequentiality is a step towards the resolution of complex robotic tasks under minimal specification effort, a key feature for the next generation of autonomous robots.

The development of machine learning algorithms has been gathering relevance to address the increasing modelling complexity of manufacturing decision-making problems. Reinforcement learning is a methodology with great potential due to the reduced need for previous training data, i.e., the system learns along time with actual operation. This study focuses on the implementation of a reinforcement learning algorithm in an assembly problem of a given object, aiming to identify the effectiveness of the proposed approach in the optimisation of the assembly process time. A model-free Q-Learning algorithm is applied, considering the learning of a matrix of Q-values (Q-table) from the successive interactions with the environment to suggest an assembly sequence solution. This implementation explores three scenarios with increasing complexity so that the impact of the Q-Learning\textsc's parameters and rewards is assessed to improve the reinforcement learning agent performance. The optimisation approach achieved very promising results by learning the optimal assembly sequence 98.3% of the times.

Jumping is essential for legged robots to traverse through difficult terrains. In this work, we propose a hierarchical framework that combines optimal control and reinforcement learning to learn continuous jumping motions for quadrupedal robots. The core of our framework is a stance controller, which combines a manually designed acceleration controller with a learned residual policy. As the acceleration controller warm starts policy for efficient training, the trained policy overcomes the limitation of the acceleration controller and improves the jumping stability. In addition, a low-level whole-body controller converts the body pose command from the stance controller to motor commands. After training in simulation, our framework can be deployed directly to the real robot, and perform versatile, continuous jumping motions, including omni-directional jumps at up to 50cm high, 60cm forward, and jump-turning at up to 90 degrees. Please visit our website for more results: https://sites.google.com/view/learning-to-jump.

The operation of electricity grids has become increasingly complex due to the current upheaval and the increase in renewable energy production. As a consequence, active grid management is reaching its limits with conventional approaches. In the context of the Learning to Run a Power Network challenge, it has been shown that Reinforcement Learning (RL) is an efficient and reliable approach with considerable potential for automatic grid operation. In this article, we analyse the submitted agent from Binbinchen and provide novel strategies to improve the agent, both for the RL and the rule-based approach. The main improvement is a N-1 strategy, where we consider topology actions that keep the grid stable, even if one line is disconnected. More, we also propose a topology reversion to the original grid, which proved to be beneficial. The improvements are tested against reference approaches on the challenge test sets and are able to increase the performance of the rule-based agent by 27%. In direct comparison between rule-based and RL agent we find similar performance. However, the RL agent has a clear computational advantage. We also analyse the behaviour in an exemplary case in more detail to provide additional insights. Here, we observe that through the N-1 strategy, the actions of the agents become more diversified.

Nowadays, the mainstream approach in position allocation system is to utilize a reinforcement learning model to allocate appropriate locations for items in various channels and then mix them into the feed. There are two types of data employed to train reinforcement learning (RL) model for position allocation, named strategy data and random data. Strategy data is collected from the current online model, it suffers from an imbalanced distribution of state-action pairs, resulting in severe overestimation problems during training. On the other hand, random data offers a more uniform distribution of state-action pairs, but is challenging to obtain in industrial scenarios as it could negatively impact platform revenue and user experience due to random exploration. As the two types of data have different distributions, designing an effective strategy to leverage both types of data to enhance the efficacy of the RL model training has become a highly challenging problem. In this study, we propose a framework named Multi-Distribution Data Learning (MDDL) to address the challenge of effectively utilizing both strategy and random data for training RL models on mixed multi-distribution data. Specifically, MDDL incorporates a novel imitation learning signal to mitigate overestimation problems in strategy data and maximizes the RL signal for random data to facilitate effective learning. In our experiments, we evaluated the proposed MDDL framework in a real-world position allocation system and demonstrated its superior performance compared to the previous baseline. MDDL has been fully deployed on the Meituan food delivery platform and currently serves over 300 million users.

The potential of offline reinforcement learning (RL) is that high-capacity models trained on large, heterogeneous datasets can lead to agents that generalize broadly, analogously to similar advances in vision and NLP. However, recent works argue that offline RL methods encounter unique challenges to scaling up model capacity. Drawing on the learnings from these works, we re-examine previous design choices and find that with appropriate choices: ResNets, cross-entropy based distributional backups, and feature normalization, offline Q-learning algorithms exhibit strong performance that scales with model capacity. Using multi-task Atari as a testbed for scaling and generalization, we train a single policy on 40 games with near-human performance using up-to 80 million parameter networks, finding that model performance scales favorably with capacity. In contrast to prior work, we extrapolate beyond dataset performance even when trained entirely on a large (400M transitions) but highly suboptimal dataset (51% human-level performance). Compared to return-conditioned supervised approaches, offline Q-learning scales similarly with model capacity and has better performance, especially when the dataset is suboptimal. Finally, we show that offline Q-learning with a diverse dataset is sufficient to learn powerful representations that facilitate rapid transfer to novel games and fast online learning on new variations of a training game, improving over existing state-of-the-art representation learning approaches.

Machine learning algorithms have achieved superhuman performance in specific complex domains. Yet learning online from few examples and efficiently generalizing across domains remains elusive. In humans such learning proceeds via declarative memory formation and is closely associated with consciousness. Predictive processing has been advanced as a principled Bayesian inference framework for understanding the cortex as implementing deep generative perceptual models for both sensory data and action control. However, predictive processing offers little direct insight into fast compositional learning or the mystery of consciousness. Here we propose that through implementing online learning by hierarchical binding of unpredicted inferences, a predictive processing system may flexibly generalize in novel situations by forming working memories for perceptions and actions from single examples, which can become short- and long-term declarative memories retrievable by associative recall. We argue that the contents of such working memories are unified yet differentiated, can be maintained by selective attention and are consistent with observations of masking, postdictive perceptual integration, and other paradigm cases of consciousness research. We describe how the brain could have evolved to use perceptual value prediction for reinforcement learning of complex action policies simultaneously implementing multiple survival and reproduction strategies. 'Conscious experience' is how such a learning system perceptually represents its own functioning, suggesting an answer to the meta problem of consciousness. Our proposal naturally unifies feature binding, recurrent processing, and predictive processing with global workspace, and, to a lesser extent, the higher order theories of consciousness.

Building a robot that can understand and learn to interact by watching humans has inspired several vision problems. However, despite some successful results on static datasets, it remains unclear how current models can be used on a robot directly. In this paper, we aim to bridge this gap by leveraging videos of human interactions in an environment centric manner. Utilizing internet videos of human behavior, we train a visual affordance model that estimates where and how in the scene a human is likely to interact. The structure of these behavioral affordances directly enables the robot to perform many complex tasks. We show how to seamlessly integrate our affordance model with four robot learning paradigms including offline imitation learning, exploration, goal-conditioned learning, and action parameterization for reinforcement learning. We show the efficacy of our approach, which we call VRB, across 4 real world environments, over 10 different tasks, and 2 robotic platforms operating in the wild. Results, visualizations and videos at https://robo-affordances.github.io/

An unmanned surface vehicle (USV) can perform complex missions by continuously observing the state of its surroundings and taking action toward a goal. A SWARM of USVs working together can complete missions faster, and more effectively than a single USV alone. In this paper, we propose an autonomous communication model for a swarm of USVs. The goal of this system is to implement a software system using Robot Operating System (ROS) and Gazebo. With the main objective of coordinated task completion, the Markov decision process (MDP) provides a base to formulate a task decision problem to achieve efficient localization and tracking in a highly dynamic water environment. To coordinate multiple USVs performing real-time target tracking, we propose an enhanced multi-agent reinforcement learning approach. Our proposed scheme uses MA-DDPG, or Multi-Agent Deep Deterministic Policy Gradient, an extension of the Deep Deterministic Policy Gradients (DDPG) algorithm that allows for decentralized control of multiple agents in a cooperative environment. MA-DDPG's decentralised control allows each and every agent to make decisions based on its own observations and objectives, which can lead to superior gross performance and improved stability. Additionally, it provides communication and coordination among agents through the use of collective readings and rewards.