Keming Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Keming Wang has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Bin Jiang1 shared publication
- Chengli Wang
Zhejiang Scientific Research Institute of Transport,Hangzhou,Zhejiang
1 shared publication - Colin Snook
University of Southampton
1 shared publication - Dana Dghaym
Department of Computer Science and Engineering, American University of Sharjah, Sharjah, United Arab Emirates
1 shared publication Fan Yang
1 shared publicationGuan‐Feng Wu
National‐Local Joint Engineering Laboratory of System Credibility Automatic Verification Southwest Jiaotong University Chengdu China
1 shared publication
Latest Publications
- Open Access
- 29 September 2026
- Published by Elsevier in SSRN Electronic Journal
ESE-HRL: Event-Triggered SafetyEciency Hierarchical Reinforcement Learning for MaplessNavigation in Unknown Dynamic Environments
Mapless navigation in unknown dynamic environments requires mobile robots to coordinate long-range goal progression, dynamic risk response, and realtime motion control despite limited local perception, unknown environmental structure, and uncertain obstacle motion. Existing methods often couple direction selection with continuous control in a single policy or update subgoals at xed intervals, making it dicult to balance timely responses,local safety, and navigation eciency. We propose an event-triggered safety eciency hierarchical reinforcement learning method (ESE-HRL) comprising a high-level subgoal planner and a goal-conditioned low-level continuous controller, which update local directions on demand and generate continuous motion commands, respectively. ESE-HRL uses only a short history of light detection and ranging (LiDAR) scans to represent local dynamics and estimate risk, supporting on-demand high-level updates and reducing interference between high-level direction selection and real-time obstacle avoidance.It constructs a nite set of candidate subgoals from locally traversable geometry and selects the motion direction by jointly evaluating eciency value and safety cost, thereby accounting for execution feasibility, goal-directed progress, and dynamic safety. The low-level policy continuously tracks the selectedsubgoal while avoiding dynamic obstacles. Experimental results show that ESE-HRL outperforms multiple learning-based baselines in success rate,safety, eciency, and generalization, and achieves lower collision rates than the Contextual-Memory Oriented Reinforcement-Learning (CORE) Planner across all test environments. These results indicate that the method can control dynamic collision risk while maintaining progress toward the goal and can adapt to unseen scenarios and variations in obstacle density.
- 29 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessESE-HRL: Event-Triggered SafetyEciency Hierarchical Reinforcement Learning for MaplessNavigation in Unknown Dynamic Environments
Mapless navigation in unknown dynamic environments requires mobile robots to coordinate long-range goal progression, dynamic risk response, and realtime motion control despite limited local perception, unknown environmental structure, and uncertain obstacle motion. Existing methods often couple direction selection with continuous control in a single policy or update subgoals at xed intervals, making it dicult to balance timely responses,local safety, and navigation eciency. We propose an event-triggered safety eciency hierarchical reinforcement learning method (ESE-HRL) comprising a high-level subgoal planner and a goal-conditioned low-level continuous controller, which update local directions on demand and generate continuous motion commands, respectively. ESE-HRL uses only a short history of light detection and ranging (LiDAR) scans to represent local dynamics and estimate risk, supporting on-demand high-level updates and reducing interference between high-level direction selection and real-time obstacle avoidance.It constructs a nite set of candidate subgoals from locally traversable geometry and selects the motion direction by jointly evaluating eciency value and safety cost, thereby accounting for execution feasibility, goal-directed progress, and dynamic safety. The low-level policy continuously tracks the selectedsubgoal while avoiding dynamic obstacles. Experimental results show that ESE-HRL outperforms multiple learning-based baselines in success rate,safety, eciency, and generalization, and achieves lower collision rates than the Contextual-Memory Oriented Reinforcement-Learning (CORE) Planner across all test environments. These results indicate that the method can control dynamic collision risk while maintaining progress toward the goal and can adapt to unseen scenarios and variations in obstacle density.
- Open Access
- 19 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 15-21
A Logic-Driven Workflow Based on LLM Agents for High-Fidelity SMT Code Generation
- 19 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 15-21
Open AccessA Logic-Driven Workflow Based on LLM Agents for High-Fidelity SMT Code Generation
- Open Access
A SCADE Model Verification Method Based on B-Model Transformation
Due to the limitations of SCADE models in expressing and verifying abstract specifications in safety-critical systems, this study proposes a formal verification framework based on the B-Method. By establishing a semantic equivalence transformation mechanism from SCADE models to B models, a hierarchical mapping rule set is constructed, covering type systems, control flow structures, and state machines. This effectively addresses key technical challenges such as loop-equivalent transformation proof for high-order operators and modeling of temporal logic storage structures. The proposed method innovatively leverages the abstraction capabilities of B-Method in set theory and first-order logic, overcoming the constraints of native verification tools of SCADE in complex specification descriptions. It successfully verifies abstract specifications that are difficult to model directly in SCADE. Experimental results show that the transformed B models achieve a higher defect detection rate and improved verification efficiency in the ProB verification environment compared to the native verifier of SCADE, significantly enhancing the formal verification capability of safety-critical systems. This study provides a cross-model verification paradigm for embedded control systems in avionics, rail transportation, and other domains, demonstrating substantial engineering application value.
Open AccessA SCADE Model Verification Method Based on B-Model Transformation
Due to the limitations of SCADE models in expressing and verifying abstract specifications in safety-critical systems, this study proposes a formal verification framework based on the B-Method. By establishing a semantic equivalence transformation mechanism from SCADE models to B models, a hierarchical mapping rule set is constructed, covering type systems, control flow structures, and state machines. This effectively addresses key technical challenges such as loop-equivalent transformation proof for high-order operators and modeling of temporal logic storage structures. The proposed method innovatively leverages the abstraction capabilities of B-Method in set theory and first-order logic, overcoming the constraints of native verification tools of SCADE in complex specification descriptions. It successfully verifies abstract specifications that are difficult to model directly in SCADE. Experimental results show that the transformed B models achieve a higher defect detection rate and improved verification efficiency in the ProB verification environment compared to the native verifier of SCADE, significantly enhancing the formal verification capability of safety-critical systems. This study provides a cross-model verification paradigm for embedded control systems in avionics, rail transportation, and other domains, demonstrating substantial engineering application value.
- 1 September 2023
- Published by American Scientific Publishers in Materials Express
- Vol. 13 (9) , 1597-1604
Detection and remediation of heavy metal pollution in seawater using instrumentation and nanomaterials
This study explores methodologies for removing heavy metal elements such as nickel (Ni), copper (Cu), cadmium (Cd), and lead (Pb) from diverse aquatic environments, including rivers, lakes, and oceans. Nanosized montmorillonite (MON) was used as the raw material and was subjected to organic chemical modification through silanization using cetyltrimethylammonium bromide and grafting of amino groups to produce amino-functionalized nanomontmorillonite composite (NH2-MON). The removal effectiveness of NH2-MON on heavy metal elements in water bodies was evaluated. Experiments involving adsorption were conducted to evaluate the impact of nanomaterial concentration and solution pH on the entrapment of heavy metal ions. The results indicated that an increased nanomaterial adsorbent dosage precipitated water coagulation, which subsequently altered the accessibility of adsorption sites for heavy metal ions, thereby significantly affecting the heavy metal removal effectiveness of the nanomaterial. The ideal nanomaterial dosage was determined to be 2.5 g/L, yielding the maximum unit adsorption capacity and removal rate. The acidity or alkalinity of the solution was instrumental in the adsorption of heavy metal ions such as Ni, Cu, Cd, and Pb using nanomaterials, establishing solution pH as a pivotal determinant in the adsorption process. As the solution pH increased, the electronegativity of the nanomaterial increased, thus encouraging its interaction with positively charged heavy metal ions, including Ni, Cu, Cd, and Pb. The ideal solution pH range was found to be 4–5.
- 1 September 2023
- Published by American Scientific Publishers in Materials Express
- Vol. 13 (9) , 1597-1604
Detection and remediation of heavy metal pollution in seawater using instrumentation and nanomaterials
This study explores methodologies for removing heavy metal elements such as nickel (Ni), copper (Cu), cadmium (Cd), and lead (Pb) from diverse aquatic environments, including rivers, lakes, and oceans. Nanosized montmorillonite (MON) was used as the raw material and was subjected to organic chemical modification through silanization using cetyltrimethylammonium bromide and grafting of amino groups to produce amino-functionalized nanomontmorillonite composite (NH2-MON). The removal effectiveness of NH2-MON on heavy metal elements in water bodies was evaluated. Experiments involving adsorption were conducted to evaluate the impact of nanomaterial concentration and solution pH on the entrapment of heavy metal ions. The results indicated that an increased nanomaterial adsorbent dosage precipitated water coagulation, which subsequently altered the accessibility of adsorption sites for heavy metal ions, thereby significantly affecting the heavy metal removal effectiveness of the nanomaterial. The ideal nanomaterial dosage was determined to be 2.5 g/L, yielding the maximum unit adsorption capacity and removal rate. The acidity or alkalinity of the solution was instrumental in the adsorption of heavy metal ions such as Ni, Cu, Cd, and Pb using nanomaterials, establishing solution pH as a pivotal determinant in the adsorption process. As the solution pH increased, the electronegativity of the nanomaterial increased, thus encouraging its interaction with positively charged heavy metal ions, including Ni, Cu, Cd, and Pb. The ideal solution pH range was found to be 4–5.
- Open Access
- 21 April 2022
- Published by Institution of Engineering and Technology (IET) in IET Intelligent Transport Systems
- Vol. 16 (8) , 1042-1057
Logical consistency verification of state sensing in safety‐critical decision: A case study of train routing selection
Reliable decision‐making in a safety‐critical system depends on the consistency of the state of the system process. Based on the decision‐makeing properties of safety‐critical systems, the definition of the critical component, Boolean evolution system, and cognate variables are presented here. By exploring the two lemmas of the critical components, the consistency of the safety‐critical system is defined, then a new process of a safety‐critical decision of consistency checking is proposed. In order to check the consistency properties of the decision, two computation tree logic formulas derived from the lemmas of the consistency checking are developed. The case of the train routing selection is modelled and verified by model checking, which shows the decision prototype guarantees the safety of the critical process. As the turnout is the critical component in the process of the train routing selection, a multi‐sensor‐based unit of state monitoring is developed, and a real‐world test shows the accuracy of the monitoring unit. This study confirms that the state consistency verification ensures the reliability of the safety‐critical system's decision output. The case can enlighten the development of the intelligent control system which has a critical component of decision‐making.
- 21 April 2022
- Published by Institution of Engineering and Technology (IET) in IET Intelligent Transport Systems
- Vol. 16 (8) , 1042-1057
Open AccessLogical consistency verification of state sensing in safety‐critical decision: A case study of train routing selection
Reliable decision‐making in a safety‐critical system depends on the consistency of the state of the system process. Based on the decision‐makeing properties of safety‐critical systems, the definition of the critical component, Boolean evolution system, and cognate variables are presented here. By exploring the two lemmas of the critical components, the consistency of the safety‐critical system is defined, then a new process of a safety‐critical decision of consistency checking is proposed. In order to check the consistency properties of the decision, two computation tree logic formulas derived from the lemmas of the consistency checking are developed. The case of the train routing selection is modelled and verified by model checking, which shows the decision prototype guarantees the safety of the critical process. As the turnout is the critical component in the process of the train routing selection, a multi‐sensor‐based unit of state monitoring is developed, and a real‐world test shows the accuracy of the monitoring unit. This study confirms that the state consistency verification ensures the reliability of the safety‐critical system's decision output. The case can enlighten the development of the intelligent control system which has a critical component of decision‐making.
- Open Access
- 27 December 2018
- Published by Springer Nature in International Journal of Disaster Risk Science
- Vol. 10 (1) , 74-88
An Emergency Blood Allocation Approach Considering Blood Group Compatibility in Disaster Relief Operations
- 27 December 2018
- Published by Springer Nature in International Journal of Disaster Risk Science
- Vol. 10 (1) , 74-88
Open AccessAn Emergency Blood Allocation Approach Considering Blood Group Compatibility in Disaster Relief Operations
- 1 August 2015
- Published by Elsevier in Transportation Research Part E: Logistics and Transportation Review
- Vol. 80, 166-183
Age-based policy for blood transshipment during blood shortage
- 1 August 2015
- Published by Elsevier in Transportation Research Part E: Logistics and Transportation Review
- Vol. 80, 166-183
Age-based policy for blood transshipment during blood shortage