Qingwei Dong

Dr. Qingwei Dong

orcid

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 November 2023 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
QDQingwei Dong
Qingwei Dong
PZPeng Zeng
Peng Zeng
GWguangxi wan
guangxi wan
YHYunpeng He
Yunpeng He
XDXiaoting Dong
Xiaoting Dong
Deepreinforcement learning algorithms offer a promising method for industrial robots to tackle unstructured and complex scenarios that are difficult to model. However, due to constraints related to equipment lifespan and safety requirements, acquiring a number of samples directly from the physical environment is often infeasible. With the development of increasingly realistic simulators, it has become feasible for industrial robots to acquire complex motion skills within simulated environments. Nonetheless, the “reality gap” frequently results in performance degradation when transferring policies trained in simulators to physical systems. In this letter, we treat the reality gap between a physical environment (target domain) and a simulated environment (source domain) as a Gaussian perturbation and utilize Kalman filtering to reduce the discrepancy between source and target domain data. We refine the source domain controller using target domain data to enhance the controller's adaptability to the target domain. The efficacy of the proposed method is demonstrated in reaching tasks and peg-in-hole tasks conducted on PR2 and UR5 robotic platforms.

29 October 2024 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
QDQingwei Dong
Qingwei Dong
PZPeng Zeng
Peng Zeng
YHYunpeng He
Yunpeng He
GWguangxi wan
guangxi wan
XDXiaoting Dong
Xiaoting Dong
Complex operational scenarios increasingly demand that industrial robots sequentially resolve multiple interrelated problems to accomplish complex operational tasks, necessitating robots to have the capacity for not only learning through interaction with the environment but also for continual learning. Current deep reinforcement learning methods have demonstrated substantial prowess in enabling robots to learn individual simple operational skills. However, catastrophic forgetting regarding the continual learning of various distinct tasks under a unified control policy remains a challenge. The lengthy sequential decision-making trajectory in reinforcement learning scenarios results in a massive state-action search space for the agent. Moreover, low-value state-action samples exacerbate the difficulty of continuous learning in reinforcement learning problems. In this letter, we propose a Continual Reinforcement Learning (CRL) method that accommodates the incremental multiskill learning demands of robots. We transform the tightly coupled structure in Guided Policy Search (GPS) algorithms, which closely intertwine local and global policies, into a loosely coupled structure. This revised structure updates the global policy only after the local policy for a specific task has converged, enabling online learning. In incrementally learning new tasks, the global policy is updated using hard parameter sharing and Memory Aware Synapses (MAS), creating task-specific layers while penalizing significant parameter changes in shared layers linked to prior tasks. This method reduces overfitting and mitigates catastrophic forgetting in robotic CRL. We validate our method on PR2, UR5 and Sawyer robots in simulators as well as on a real UR5 robot.

19 March 2025 in Sustainability
Verified
Journal Article
Open access
PDF file
XWXiaohui Wang
Xiaohui Wang
SCShijie Cui
Shijie Cui
QDQingwei Dong
Qingwei Dong
Driven by policy incentives and economic pressures, energy-intensive industries are increasingly focusing on energy cost reductions amid the rapid adoption of renewable energy. However, the existing studies often isolate photovoltaic-energy storage system (PV-ESS) configurations from detailed load scheduling, limiting industrial park energy management. To address this, we propose a two-layer cooperative optimization approach (TLCOA). The upper layer employs a genetic algorithm (GA) to optimize the PV capacity and energy storage sizing through natural selection and crossover operations, while the lower layer utilizes mixed integer linear programming (MILP) to derive cost-minimized scheduling strategies under time-of-use tariffs. Multi-process parallel computing accelerates the fitness evaluations, resolving high-dimensional industrial data challenges. Multi-process parallel computing is introduced to accelerate fitness evaluations, effectively addressing the challenges posed by high-dimensional industrial data. Validated with real power market data, the TLCOA demonstrated rapid adaptation to load fluctuations while achieving a 23.68% improvement in computational efficiency, 1.73% reduction in investment costs, 7.55% decrease in power purchase costs, and 8.79% enhancement in renewable energy utilization compared to traditional methods. This integrated framework enables cost-effective PV-ESS deployment and adaptive energy management in industrial facilities, offering actionable insights for renewable integration and scalable energy optimization.

7 April 2025 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
QDQingwei Dong
Qingwei Dong
TWTingting Wu
Tingting Wu
PZPeng Zeng
Peng Zeng
CZChuanzhi Zang
Chuanzhi Zang
GWguangxi wan
guangxi wan
SCShijie Cui
Shijie Cui
In the domain of robot learning, equipping robots with the capability to swiftly acquire operational skills poses a significant challenge. Currently, reinforcement learning techniques are adept at addressing dynamic, unstructured problems involving rich contact scenarios. However, the convergence rate of these algorithms is often slow due to the high dimensionality of the robot state-action mapping space and the extensive initial policy search space. Meanwhile, advancements in large language models (LLMs) have endowed these models with a degree of logical reasoning ability, enabling them to take goal-oriented actions proactively during the initial phase of a robotic task. These models can implicitly generate features of states and uncover underlying patterns in trajectory generation. Yet, in complex manipulative tasks involving rich contact scenarios, LLMs still fall short. Thus, integrating the robust interactive capabilities of reinforcement learning with the strong logical reasoning of LLMs, and enhancing policy search with LLMs, could potentially accelerate the speed of policy searches significantly. In this paper, we introduce a Cognitive Reasoning Trajectory Optimization method. This approach utilizes Low-level Cognitive Control Tuning to enable LLMs with robust logical reasoning to make effective single-step decisions in Markov Decision Process (MDP) tasks. By fitting dynamic models with high-quality cognitive reasoning data and optimizing control strategies, this method constrains the policy search space and enhances the efficiency of trajectory optimization. Experimental results on various manipulative tasks using the Sawyer robot in the Mujoco simulator validate the effectiveness of the proposed algorithm.

31 January 2026 in Complex & Intelligent Systems
Verified
Journal Article
Open access
PDF file
TWTingting Wu
Tingting Wu
QDQingwei Dong
Qingwei Dong
WYWei Yu
Wei Yu
ZLZheng Li
Zheng Li
MGManxue Guo
Manxue Guo
Learning rate is the most important parameter for neural network performance optimization, and its selection and theoretical exploration have always been an active research field. Although the learning rate scheduler has been widely observed to improve the performance of model training, the current works do not consider the characteristics of different stages of training and are mostly incompatible with adaptive optimizers. In this work, we construct a general learning rate scheduler based on the different stages of the training process "warm-up - maintain large learning rate - decay", which can meet the changes of learning rate in different training stages while reducing the dependence on hyperparameters. In addition, we theoretically demonstrate the convergence of the proposed general learning rate scheduler on the adaptive optimizers. Compared with the mainstream learning rate scheduler on the current common data sets, we comprehensively evaluate the advanced and robust performance of the proposed learning rate scheduler through a large number of experiments.