Robotics has gained significant attention due to its autonomy and ability to automate in the nuclear industry. However, the increasing complexity of robots has led to a growing demand for advanced simulation and control methods to predict robot behavior and optimize plant performance. Most existing digital twins only address parts of systems and do not offer an overall design of nuclear power plants. Furthermore, they are often designed for specific algorithms or tasks, making them unsuitable for broader research applications or other potential projects. In response, we propose a comprehensive nuclear power plant designed to enhance real-time monitoring, operational efficiency, and predictive maintenance. We selected to model a full-scope nuclear power plant in Unreal Engine 5 to incorporate the complexities and various phenomena. The high-resolution simulation environment is integrated with a General Pressurized Water Reactor Simulator, a high-fidelity physics-driven software, to create a realistic flow of nuclear power plant and a real-time updating virtual environment. Furthermore, the virtual environment provides various features and a Python bridge for researchers to test custom algorithms and frameworks easily. The digital twin's performance is presented, and several research ideas - such as multi-robot task scheduling and robot navigation in the radiation area - using implemented features are presented.

Accurate physical simulation is crucial for the development and validation of control algorithms in robotic systems. Recent works in Reinforcement Learning (RL) take notably advantage of extensive simulations to produce efficient robot control. State-of-the-art servo actuator models generally fail at capturing the complex friction dynamics of these systems. This limits the transferability of simulated behaviors to real-world applications. In this work, we present extended friction models that allow to more accurately simulate servo actuator dynamics. We propose a comprehensive analysis of various friction models, present a method for identifying model parameters using recorded trajectories from a pendulum test bench, and demonstrate how these models can be integrated into physics engines. The proposed friction models are validated on four distinct servo actuators and tested on 2R manipulators, showing significant improvements in accuracy over the standard Coulomb-Viscous model. Our results highlight the importance of considering advanced friction effects in the simulation of servo actuators to enhance the realism and reliability of robotic simulations.

Reinforcement Learning (RL) plays a crucial role in aligning large language models (LLMs) with human preferences and improving their ability to perform complex tasks. However, current approaches either require significant computational resources due to the use of multiple models and extensive online sampling for training (e.g., PPO) or are framed as bandit problems (e.g., DPO, DRO), which often struggle with multi-step reasoning tasks, such as math problem-solving and complex reasoning that involve long chains of thought. To overcome these limitations, we introduce Direct Q-function Optimization (DQO), which formulates the response generation process as a Markov Decision Process (MDP) and utilizes the soft actor-critic (SAC) framework to optimize a Q-function directly parameterized by the language model. The MDP formulation of DQO offers structural advantages over bandit-based methods, enabling more effective process supervision. Experimental results on two math problem-solving datasets, GSM8K and MATH, demonstrate that DQO outperforms previous methods, establishing it as a promising offline reinforcement learning approach for aligning language models.

Planning and performing interactive tasks, such as conducting experiments to determine the melting point of an unknown substance, is straightforward for humans but poses significant challenges for autonomous agents. We introduce ReasonPlanner, a novel generalist agent designed for reflective thinking, planning, and interactive reasoning. This agent leverages LLMs to plan hypothetical trajectories by building a World Model based on a Temporal Knowledge Graph. The agent interacts with the environment using a natural language actor-critic module, where the actor translates the imagined trajectory into a sequence of actionable steps, and the critic determines if replanning is necessary. ReasonPlanner significantly outperforms previous state-of-the-art prompting-based methods on the ScienceWorld benchmark by more than 1.8 times, while being more sample-efficient and interpretable. It relies solely on frozen weights thus requiring no gradient updates. ReasonPlanner can be deployed and utilized without specialized knowledge of Machine Learning, making it accessible to a wide range of users.

Model-based reinforcement learning (RL) offers a solution to the data inefficiency that plagues most model-free RL algorithms. However, learning a robust world model often demands complex and deep architectures, which are expensive to compute and train. Within the world model, dynamics models are particularly crucial for accurate predictions, and various dynamics-model architectures have been explored, each with its own set of challenges. Currently, recurrent neural network (RNN) based world models face issues such as vanishing gradients and difficulty in capturing long-term dependencies effectively. In contrast, use of transformers suffers from the well-known issues of self-attention mechanisms, where both memory and computational complexity scale as $O(n^2)$, with $n$ representing the sequence length. To address these challenges we propose a state space model (SSM) based world model, specifically based on Mamba, that achieves $O(n)$ memory and computational complexity while effectively capturing long-term dependencies and facilitating the use of longer training sequences efficiently. We also introduce a novel sampling method to mitigate the suboptimality caused by an incorrect world model in the early stages of training, combining it with the aforementioned technique to achieve a normalised score comparable to other state-of-the-art model-based RL algorithms using only a 7 million trainable parameter world model. This model is accessible and can be trained on an off-the-shelf laptop. Our code is available at https://github.com/realwenlongwang/drama.git.

Designing Public Transport (PT) networks able to satisfy mobility needs of people is essential to reduce the number of individual vehicles on the road, and thus pollution and congestion. Urban sustainability is thus tightly coupled to an efficient PT. Current approaches on Transport Network Design (TND) generally aim to optimize generalized cost, i.e., a unique number including operator and users' costs. Since we intend quality of PT as the capability of satisfying mobility needs, we focus instead on PT accessibility, i.e., the ease of reaching surrounding points of interest via PT. PT accessibility is generally unequally distributed in urban regions: suburbs generally suffer from poor PT accessibility, which condemns residents therein to be dependent on their private cars. We thus tackle the problem of designing bus lines so as to minimize the inequality in the geographical distribution of accessibility. We combine state-of-the-art Message Passing Neural Networks (MPNN) and Reinforcement Learning. We show the efficacy of our method against metaheuristics (classically used in TND) in a use case representing in simplified terms the city of Montreal.

In this paper, we establish the global convergence of the actor-critic algorithm with a significantly improved sample complexity of $O(\epsilon^{-3})$, advancing beyond the existing local convergence results. Previous works provide local convergence guarantees with a sample complexity of $O(\epsilon^{-2})$ for bounding the squared gradient of the return, which translates to a global sample complexity of $O(\epsilon^{-4})$ using the gradient domination lemma. In contrast to traditional methods that employ decreasing step sizes for both the actor and critic, we demonstrate that a constant step size for the critic is sufficient to ensure convergence in expectation. This key insight reveals that using a decreasing step size for the actor alone is sufficient to handle the noise for both the actor and critic. Our findings provide theoretical support for the practical success of many algorithms that rely on constant step sizes.

The predominant approach in reinforcement learning is to assign credit to actions based on the expected return. However, we show that the return may depend on the policy in a way which could lead to excessive variance in value estimation and slow down learning. Instead, we show that the advantage function can be interpreted as causal effects and shares similar properties with causal representations. Based on this insight, we propose Direct Advantage Estimation (DAE), a novel method that can model the advantage function and estimate it directly from on-policy data while simultaneously minimizing the variance of the return without requiring the (action-)value function. We also relate our method to Temporal Difference methods by showing how value functions can be seamlessly integrated into DAE.

Model-based reinforcement learning algorithms with probabilistic dynamical models are amongst the most data-efficient learning methods. This is often attributed to their ability to distinguish between epistemic and aleatoric uncertainty. However, while most algorithms distinguish these two uncertainties for learning the model, they ignore it when optimizing the policy, which leads to greedy and insufficient exploration. At the same time, there are no practical solvers for optimistic exploration algorithms. In this paper, we propose a practical optimistic exploration algorithm (H-UCRL).

How to select between policies and value functions produced by different training algorithms in offline reinforcement learning (RL)---which is crucial for hyperparameter tuning---is an important open question. Existing approaches based on off-policy evaluation (OPE) often require additional function approximation and hence hyperparameters, creating a chicken-and-egg situation. In this paper, we design hyperparameter-free algorithms for policy selection based on BVFT [XJ21], a recent theoretical advance in value-function selection, and demonstrate their effectiveness in discrete-action benchmarks such as Atari. To address performance degradation due to poor critics in continuous-action domains, we further combine BVFT with OPE to get the best of both worlds, and obtain a hyperparameter-tuning method for Q -function based OPE with theoretical guarantees as a side product.