This work explores techniques to scale up image-based end-to-end learning for dexterous grasping with an arm + hand system. Unlike state-based RL, vision-based RL is much more memory inefficient, resulting in relatively low batch sizes, which is not amenable for algorithms like PPO. Nevertheless, it is still an attractive method as unlike the more commonly used techniques which distill state-based policies into vision networks, end-to-end RL can allow for emergent active vision behaviors. We identify a key bottleneck in training these policies is the way most existing simulators scale to multiple GPUs using traditional data parallelism techniques. We propose a new method where we disaggregate the simulator and RL (both training and experience buffers) onto separate GPUs. On a node with four GPUs, we have the simulator running on three of them, and PPO running on the fourth. We are able to show that with the same number of GPUs, we can double the number of existing environments compared to the previous baseline of standard data parallelism. This allows us to train vision-based environments, end-to-end with depth, which were previously performing far worse with the baseline. We train and distill both depth and state-based policies into stereo RGB networks and show that depth distillation leads to better results, both in simulation and reality. This improvement is likely due to the observability gap between state and vision policies which does not exist when distilling depth policies into stereo RGB. We further show that the increased batch size brought about by disaggregated simulation also improves real world performance. When deploying in the real world, we improve upon the previous state-of-the-art vision-based results using our end-to-end policies.

Continually adapting edge models in cloud-edge collaborative object detection for traffic monitoring suffers from catastrophic forgetting, where models lose previously learned knowledge when adapting to new data distributions. This is especially problematic in dynamic traffic environments characterised by periodic variations (e.g., day/night, peak hours), where past knowledge remains valuable. Existing approaches like experience replay and visual prompts offer some mitigation, but struggle to effectively prioritize and leverage historical data for optimal knowledge retention and adaptation. Specifically, simply storing and replaying all historical data can be inefficient, while treating all historical experiences as equally important overlooks their varying relevance to the current domain. This paper proposes ER-EMU, an edge model update algorithm based on adaptive experience replay, to address these limitations. ER-EMU utilizes a limited-size experience buffer managed using a First-In-First-Out (FIFO) principle, and a novel Domain Distance Metric-based Experience Selection (DDM-ES) algorithm. DDM-ES employs the multi-kernel maximum mean discrepancy (MK-MMD) to quantify the dissimilarity between target domains, prioritizing the selection of historical data that is most dissimilar to the current target domain. This ensures training diversity and facilitates the retention of knowledge from a wider range of past experiences, while also preventing overfitting to the new domain. The experience buffer is also updated using a simple random sampling strategy to maintain a balanced representation of previous domains. Experiments on the Bellevue traffic video dataset, involving repeated day/night cycles, demonstrate that ER-EMU consistently improves the performance of several state-of-the-art cloud-edge collaborative object detection frameworks.

The exact mechanic of the policy transfer between different algorithm is not given. Given the content, I may assume that "transfer" means a simple copying of the parameters, but I remain unsure. When augmenting the experience buffer with other algorithm, it would be nice to clarify why it does (not) introduce any bias in the data. It seems that the different parts of the framework could be replaced by a different way of "tinkering" with a algorithm or its hyperparameters. E.g., the auxiliary on-policy algorithms are here mainly for exploration, but the exploration of the main off-policy algorithm itself can be easily controlled and I suspect it can, with the right setting, work as good as the given complicated framework. The global and local experience buffer seems more like a hack.

Mathematical equations have been unreasonably effective in describing complex natural phenomena across various scientific disciplines. However, discovering such insightful equations from data presents significant challenges due to the necessity of navigating extremely high-dimensional combinatorial and nonlinear hypothesis spaces. Traditional methods of equation discovery, commonly known as symbolic regression, largely focus on extracting equations from data alone, often neglecting the rich domain-specific prior knowledge that scientists typically depend on. To bridge this gap, we introduce LLM-SR, a novel approach that leverages the extensive scientific knowledge and robust code generation capabilities of Large Language Models (LLMs) to discover scientific equations from data in an efficient manner. Specifically, LLM-SR treats equations as programs with mathematical operators and combines LLMs' scientific priors with evolutionary search over equation programs. The LLM iteratively proposes new equation skeleton hypotheses, drawing from its physical understanding, which are then optimized against data to estimate skeleton parameters. We demonstrate LLM-SR's effectiveness across three diverse scientific domains, where it discovers physically accurate equations that provide significantly better fits to in-domain and out-of-domain data compared to the well-established symbolic regression baselines. Incorporating scientific prior knowledge also enables LLM-SR to search the equation space more efficiently than baselines. Code is available at: https://github.com/deep-symbolic-mathematics/LLM-SR

Object rearrangement is a challenge for embodied agents because solving these tasks requires generalizing across a combinatorially large set of configurations of entities and their locations. Worse, the representations of these entities are unknown and must be inferred from sensory percepts. We present a hierarchical abstraction approach to uncover these underlying entities and achieve combinatorial generalization from unstructured visual inputs. By constructing a factorized transition graph over clusters of entity representations inferred from pixels, we show how to learn a correspondence between intervening on states of entities in the agent's model and acting on objects in the environment. We use this correspondence to develop a method for control that generalizes to different numbers and configurations of objects, which outperforms current offline deep RL methods when evaluated on simulated rearrangement tasks.

Direct Low Earth Orbit satellite-to-handheld links are expected to be part of a new era in satellite communications. Space-Division Multiple Access precoding is a technique that reduces interference among satellite beams, therefore increasing spectral efficiency by allowing cooperating satellites to reuse frequency. Over the past decades, optimal precoding solutions with perfect channel state information have been proposed for several scenarios, whereas robust precoding with only imperfect channel state information has been mostly studied for simplified models. In particular, for Low Earth Orbit satellite applications such simplified models might not be accurate. In this paper, we use the function approximation capabilities of the Soft Actor-Critic deep Reinforcement Learning algorithm to learn robust precoding with no knowledge of the system imperfections.

Inductive node-wise graph incremental learning is a challenging task due to the dynamic nature of evolving graphs and the dependencies between nodes. In this paper, we propose a novel experience replay framework, called Structure-Evolution-Aware Experience Replay (SEA-ER), that addresses these challenges by leveraging the topological awareness of GNNs and importance reweighting technique. Our framework effectively addresses the data dependency of node prediction problems in evolving graphs, with a theoretical guarantee that supports its effectiveness. Through empirical evaluation, we demonstrate that our proposed framework outperforms the current state-of-the-art GNN experience replay methods on several benchmark datasets, as measured by metrics such as accuracy and forgetting.

Human environments are often regulated by explicit and complex rulesets. Integrating Reinforcement Learning (RL) agents into such environments motivates the development of learning mechanisms that perform well in rule-dense and exception-ridden environments such as autonomous driving on regulated roads. In this paper, we propose a method for organising experience by means of partitioning the experience buffer into clusters labelled on a per-explanation basis. We present discrete and continuous navigation environments compatible with modular rulesets and 9 learning tasks. For environments with explainable rulesets, we convert rule-based explanations into case-based explanations by allocating state-transitions into clusters labelled with explanations. This allows us to sample experiences in a curricular and task-oriented manner, focusing on the rarity, importance, and meaning of events. We label this concept Explanation-Awareness (XA). We perform XA experience replay (XAER) with intra and inter-cluster prioritisation, and introduce XA-compatible versions of DQN, TD3, and SAC. Performance is consistently superior with XA versions of those algorithms, compared to traditional Prioritised Experience Replay baselines, indicating that explanation engineering can be used in lieu of reward engineering for environments with explainable features.

Expert Iteration (ExIt) is an effective framework for learning game-playing policies from self-play. ExIt involves training a policy to mimic the search behaviour of a tree search algorithm - such as Monte-Carlo tree search - and using the trained policy to guide it. The policy and the tree search can then iteratively improve each other, through experience gathered in self-play between instances of the guided tree search algorithm. This paper outlines three different approaches for manipulating the distribution of data collected from self-play, and the procedure that samples batches for learning updates from the collected data. Firstly, samples in batches are weighted based on the durations of the episodes in which they were originally experienced. Secondly, Prioritized Experience Replay is applied within the ExIt framework, to prioritise sampling experience from which we expect to obtain valuable training signals. Thirdly, a trained exploratory policy is used to diversify the trajectories experienced in self-play. This paper summarises the effects of these manipulations on training performance evaluated in fourteen different board games. We find major improvements in early training performance in some games, and minor improvements averaged over fourteen games.

A structured understanding of our world in terms of objects, relations, and hierarchies is an important component of human cognition. Learning such a structured world model from raw sensory data remains a challenge. As a step towards this goal, we introduce Contrastively-trained Structured World Models (C-SWMs). C-SWMs utilize a contrastive approach for representation learning in environments with compositional structure. We structure each state embedding as a set of object representations and their relations, modeled by a graph neural network. This allows objects to be discovered from raw pixel observations without direct supervision as part of the learning process. We evaluate C-SWMs on compositional environments involving multiple interacting objects that can be manipulated independently by an agent, simple Atari games, and a multi-object physics simulation. Our experiments demonstrate that C-SWMs can overcome limitations of models based on pixel reconstruction and outperform typical representatives of this model class in highly structured environments, while learning interpretable object-based representations.