Lidi Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Lidi Zhang has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Chao Li
Zhejiang University, Hangzhou, China
2 shared publications - Desheng Zhang
Rutgers University
2 shared publications - Haotian Wang
JD Logistics, Beijing, China
2 shared publications - Tian He
University of Minnesota
2 shared publications - Yinfeng Xiang
Zhejiang University, Hangzhou, China
2 shared publications - Yunhuai Liu
Peking University
2 shared publications
Latest Publications
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 8501-8509
CoPack: Collaborate Simulation and Reality for Robot Packing Learning in Real-world and Physical Engine
The 3D bin packing problem, with its diverse industrial applications, has garnered significant research attention in recent years. Existing approaches typically model it as a discrete and static process, while real-world applications involve continuous gravity-driven interactions. This idealized simplification leads to infeasible deployments (e.g., unstable packing) in practice. Simulations with physical engine offer an opportunity to emulate continuous gravity effects, enabling the training of reinforcement learning (RL) agents to address such limitations and improve packing stability. However, a simulation-to-reality gap persists due to dynamic variations in physical properties of real-world objects, such as various friction coefficients, elasticity, and non-uniform weight distributions. To bridge this gap, we propose a hybrid RL framework that collaborates with physical simulation with real-world data feedback. Firstly, domain randomization is applied during simulation to expose agents to a spectrum of physical parameters, enhancing their generalization capability. Secondly, the RL agent is fine-tuned with real-world deployment feedback, further reducing collapse rates. Extensive experiments demonstrate that our method achieves lower collapse rates in both simulated and real-world scenarios. Large-scale deployments in logistics systems validate the practical effectiveness, with a 35% reduction in packing collapse compared to baseline methods.
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 8501-8509
Open AccessCoPack: Collaborate Simulation and Reality for Robot Packing Learning in Real-world and Physical Engine
The 3D bin packing problem, with its diverse industrial applications, has garnered significant research attention in recent years. Existing approaches typically model it as a discrete and static process, while real-world applications involve continuous gravity-driven interactions. This idealized simplification leads to infeasible deployments (e.g., unstable packing) in practice. Simulations with physical engine offer an opportunity to emulate continuous gravity effects, enabling the training of reinforcement learning (RL) agents to address such limitations and improve packing stability. However, a simulation-to-reality gap persists due to dynamic variations in physical properties of real-world objects, such as various friction coefficients, elasticity, and non-uniform weight distributions. To bridge this gap, we propose a hybrid RL framework that collaborates with physical simulation with real-world data feedback. Firstly, domain randomization is applied during simulation to expose agents to a spectrum of physical parameters, enhancing their generalization capability. Secondly, the RL agent is fine-tuned with real-world deployment feedback, further reducing collapse rates. Extensive experiments demonstrate that our method achieves lower collapse rates in both simulated and real-world scenarios. Large-scale deployments in logistics systems validate the practical effectiveness, with a 35% reduction in packing collapse compared to baseline methods.
- 3 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 4226-4241
Uncertainty-Aware Decisioning for Robust Delivery Zone Partition in Last-Mile Logistics
In last-mile logistics, robust delivery zone partitioning is critical for ensuring timely service and maintaining a positive courier experience. However, existing network-structured spatial partitioning strategies often fail to address heterogeneous uncertainties in both delivery demand and courier supply. In particular, two challenges remain: (i) irregular demand patterns with weak periodicity, and (ii) stochastic courier availability. To this end, we propose the Prediction and Sampling Then Optimization (PSTO) paradigm, which integrates hybrid uncertainty modeling with Monte Carlo optimization. Specifically, we implement a model named RobustDZP, which combines time-series prediction with residual learning for demand estimation, a multi-agent RL sampler for supply estimation, and an iterative perturb-and-vote algorithm for partition optimization. Extensive experiments on a large real-world dataset from JD Logistics demonstrate that our approach improves delivery timeliness by over 16.3% and significantly outperforms baselines under stress tests. Furthermore, we present a real-world case study to validate the practical effectiveness of our framework and highlight key lessons for designing robust and well-received partition schemes.
- 3 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 4226-4241
Uncertainty-Aware Decisioning for Robust Delivery Zone Partition in Last-Mile Logistics
In last-mile logistics, robust delivery zone partitioning is critical for ensuring timely service and maintaining a positive courier experience. However, existing network-structured spatial partitioning strategies often fail to address heterogeneous uncertainties in both delivery demand and courier supply. In particular, two challenges remain: (i) irregular demand patterns with weak periodicity, and (ii) stochastic courier availability. To this end, we propose the Prediction and Sampling Then Optimization (PSTO) paradigm, which integrates hybrid uncertainty modeling with Monte Carlo optimization. Specifically, we implement a model named RobustDZP, which combines time-series prediction with residual learning for demand estimation, a multi-agent RL sampler for supply estimation, and an iterative perturb-and-vote algorithm for partition optimization. Extensive experiments on a large real-world dataset from JD Logistics demonstrate that our approach improves delivery timeliness by over 16.3% and significantly outperforms baselines under stress tests. Furthermore, we present a real-world case study to validate the practical effectiveness of our framework and highlight key lessons for designing robust and well-received partition schemes.
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 6266-6273
D 3 -TR: Data-driven Daily Delivery Task Rescheduling for Cost-effective Last-mile Delivery
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 6266-6273
Open AccessD 3 -TR: Data-driven Daily Delivery Task Rescheduling for Cost-effective Last-mile Delivery