LOTUSim-Energy: A Maritime Simulator for Human-Drone Interaction in Autonomous Offshore Operation \& Maintenance
本文介绍了一种名为LOTUSim-Energy的实时海上模拟器，用于解决多领域人-无人机互动下的离岸运维问题，通过统一环境物理模型、异构无人载具仿真和沉浸式监控界面实现。
0 citationsRead paper
Selected work
Representative Papers
Reasoning Errors Have a Region and a Direction in the Residual-Stream Trajectory of LLMs
This work addresses the challenge of reliably distinguishing correct from incorrect reasoning in large language models without relying on superficial shortcuts. It introduces the first approach that models reasoning errors as region- and direction-specific signals within residual streams and proposes a three-stream detector that integrates residual trajectory dynamics, vector-quantized coarse-grained regional information, and fine-grained directional cues from normalized multi-layer states to reconstruct rich contextual representations. By moving beyond methods limited to token-level shifts or single-layer probing, the proposed framework achieves up to a 12% improvement in selection accuracy over existing shift-based methods and a 21% gain over single-layer baselines on unseen reasoning benchmarks, while consistently outperforming competing probes in factual completion and verification tasks.
0 citationsRead paper
Towards Bio-inspired Heuristically Accelerated Reinforcement Learning for Adaptive Underwater Multi-Agents Behaviour
To address the challenges of high dynamic uncertainty, limited communication, and slow training convergence in underwater multi-AUV cooperative area coverage tasks, this paper proposes a PSO-guided multi-agent reinforcement learning (MARL) framework. The method integrates particle swarm optimization (PSO)—a bio-inspired metaheuristic—into the Multi-Agent Soft Actor-Critic (MSAC) algorithm, enabling heuristic-guided exploration of high-value state-action regions during early training stages and thereby improving the exploration-exploitation trade-off. Leveraging deep neural networks for function approximation and continuous-control MARL techniques, the approach significantly reduces training interaction steps and accelerates convergence to optimal collaborative policies in 2D underwater coverage simulations. Experimental results demonstrate that the proposed method achieves comparable task performance while substantially enhancing training efficiency, offering a practical pathway for deploying MARL in real-world underwater robotic systems.
0 citationsRead paper
Recent publications
Latest Papers
LOTUSim-Energy: A Maritime Simulator for Human-Drone Interaction in Autonomous Offshore Operation \& Maintenance
本文介绍了一种名为LOTUSim-Energy的实时海上模拟器，用于解决多领域人-无人机互动下的离岸运维问题，通过统一环境物理模型、异构无人载具仿真和沉浸式监控界面实现。
0 citationsRead paper
Reasoning Errors Have a Region and a Direction in the Residual-Stream Trajectory of LLMs
This work addresses the challenge of reliably distinguishing correct from incorrect reasoning in large language models without relying on superficial shortcuts. It introduces the first approach that models reasoning errors as region- and direction-specific signals within residual streams and proposes a three-stream detector that integrates residual trajectory dynamics, vector-quantized coarse-grained regional information, and fine-grained directional cues from normalized multi-layer states to reconstruct rich contextual representations. By moving beyond methods limited to token-level shifts or single-layer probing, the proposed framework achieves up to a 12% improvement in selection accuracy over existing shift-based methods and a 21% gain over single-layer baselines on unseen reasoning benchmarks, while consistently outperforming competing probes in factual completion and verification tasks.
0 citationsRead paper
Towards Bio-inspired Heuristically Accelerated Reinforcement Learning for Adaptive Underwater Multi-Agents Behaviour
To address the challenges of high dynamic uncertainty, limited communication, and slow training convergence in underwater multi-AUV cooperative area coverage tasks, this paper proposes a PSO-guided multi-agent reinforcement learning (MARL) framework. The method integrates particle swarm optimization (PSO)—a bio-inspired metaheuristic—into the Multi-Agent Soft Actor-Critic (MSAC) algorithm, enabling heuristic-guided exploration of high-value state-action regions during early training stages and thereby improving the exploration-exploitation trade-off. Leveraging deep neural networks for function approximation and continuous-control MARL techniques, the approach significantly reduces training interaction steps and accelerates convergence to optimal collaborative policies in 2D underwater coverage simulations. Experimental results demonstrate that the proposed method achieves comparable task performance while substantially enhancing training efficiency, offering a practical pathway for deploying MARL in real-world underwater robotic systems.
0 citationsRead paper