Organizations: Department of Civil and Environmental Engineering, University of California, Los Angeles, Los Angeles, CA 90095, USA · Department of Urban Planning, Tsinghua University, Beijing 100084, China · Department of Urban Planning and Design, The University of Hong Kong, Pokfulam Road, Hong Kong SAR, China · College of Urban and Environmental Sciences, Peking University, Beijing 100871, China · Department of Urban and Regional Planning, University of Florida, Gainesville, FL 32611, USA · Department of Urban Studies and Planning, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
Transit-Oriented Development (TOD) is a key strategy for advancing sustainable urban development by promoting high-density, mixed-use land development within walking distance of public transit stations. However, traditional TOD planning often relies on heuristic methods, making it difficult to efficiently identify quantitatively optimized transit-oriented land use patterns. This paper presents DRL4TOD, a deep reinforcement learning framework that automatically optimizes land use types and floor area ratios (FARs) around transit stations in alignment with TOD principles. DRL4TOD incorporates a graph neural network to capture spatial interdependencies among land parcels, road networks, and metro stations, and uses a DRL-based optimization strategy to assign land-use types and FARs to selected parcels. It jointly optimizes multiple objectives, including land use density, land use diversity, and transit capacity utilization, while minimizing reconstruction costs to account for development feasibility. Applied to two real-world station areas in Shanghai, DRL4TOD consistently outperforms rule-based approaches and genetic algorithms, achieving more effective TOD outcomes with lower reconstruction costs. By adjusting objective weights, the framework enables flexible prioritization, such as placing greater emphasis on cost minimization when financial feasibility is a concern. These results demonstrate DRL4TOD's potential as a decision-support tool for generating coordinated land-use and FAR configurations under planner-specified priorities, rather than prescribing a single fixed design. For policymakers, the framework provides parcel-level, constraint-aware alternatives that make trade-offs among competing objectives explicit and support targeted decisions about development intensity and feasibility.
Figure 2: Overview of the proposed DRL4TOD framework
Figure 3: Graph construction from spatial layout in the TOD planning context
Figure 4: Illustration of MDP transitions with spatial and graph representations
Figure 5: Spatial layouts and functional zoning of the two selected case study communities: (a) Community 01—Xintiandi Station (central urban context) and (b) Community 02—Jinjing Road Station (suburban context). Colors indicate functional land-use categories.
Land-Use Type
Maximum FAR
Hospital & Clinic
1.5
School
1.2
Recreation
2.0
Residential
2.5
Park & Open Space
0.0
Office
4.0
Table 1: Maximum allowable floor area ratio by land-use type
Station
Model
Metrics
Total Reward
LUDE
LUDI
TCU
DF
Xintiandi
Initial Plan
0.1757
0.1732
0.7858
0.1993
-
RBCP
0.2521
0.2251
0.7018
0.1728
0.3064
RBDP
0.1886
0.0753
0.7071
0.1697
0.3658
SA
0.4507
0.5751
0.8333
0.1680
0.3975
PSO
0.4542
0.6122
0.7027
0.2172
0.2935
Table 2: Quantitative performance comparison of DRL4TOD and baseline methods across two metro stations. Best results for each station are bold-underlined, and ∗ / ∗∗ denote statistical significance at p<0.1 and p<0.05 , respectively.
Figure 6: 3D FAR distribution under different planning methods at Xintiandi Station. Taller blocks represent higher FAR values.
Figure 7: 2D land use allocation patterns for Xintiandi Station under different planning methods.
Model Variant
Performance Metrics
Total Reward
LUDE
LUDI
TCU
DF
Without Metro Station Data
0.5232
0.6939
0.8365
0.2100
0.4096
MLP State Encoder
0.7984
1.3011
0.8106
0.2132
0.2701
DRL4TOD (Full Model)
1.0371
1.6965
0.8385
0.2272
0.4505
Table 3: Ablation study: performance comparison on Xintiandi Station
Figure 8: 3D visualization of the ablation study results
Figure 9: 2D visualization of the ablation study results
LUDE
LUDI
TCU
DF
Residential
Business
Office
Weight
Baseline
0.0825
0.1675
0.25
0.05
0.25
0.2
Modified LUDE
0.4050
0.0450
0.05
0.05
0.25
0.2
Value
Baseline
0.0141
0.0726
3.3397
0.8385
0.2272
0.4505
Modified LUDE
0.0452
0.0890
0.7655
0.8080
0.2653
0.3711
Table 4: Effect of residential density weight on performance metrics. The adjusted weight and its corresponding score are bold-underlined.
Figure 10: 3D spatial planning outcomes under varying TOD objective weights.
Figure 11: 2D spatial planning outcomes under varying TOD objective weights.
LUDE
LUDI
TCU
DF
Weight
Baseline
0.5
0.05
0.25
0.2
Modified LUDI
0.05
0.8
0.05
0.1
Modified TCU
0.1
0.05
0.8
0.05
Modified DF
0.1
0.05
0.05
0.8
Value
Baseline
1.6965
0.8385
0.2272
0.4505
Modified LUDI
0.3022
0.9415
0.2013
0.4876
Table 5: Effect of LUDI, TCU, and DF weights on planning performance. The adjusted weights and corresponding scores are bold-underlined.
Figure 12: 3D spatial planning outcomes under development feasibility weight adjustment. Hatched parcels indicate preserved areas.
Figure 13: 2D spatial planning outcomes under development feasibility weight adjustment. Hatched parcels indicate preserved areas.
Station
Model
Metrics
Total Reward
LUDE
LUDI
TCU
DF
Xintiandi
Weight
–
0.23
0.27
0.22
0.28
AGA
0.3865
0.1310
0.7253
0.1716
0.4386
DRL4TOD
0.4349 ∗∗
0.2232 ∗∗
0.8022 ∗∗
0.1859 ∗∗
0.4503 ∗∗
Jinjing Road
Weight
–
0.22
0.30
0.24
0.24
AGA
0.3354
0.2158
0.6154
0.1680
0.2623
Table 6: Performance of AGA and DRL4TOD under expert-elicited objective weights.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 14: XGBoost feature importance for the station-level ridership model.
Hyperparameter
Value
Metrics
Total Reward
LUDE
LUDI
TCU
DF
PPO learning rate
1×10−4
0.7756
1.1848
0.8217
0.2107
0.4472
4×10−4
1.0371
1.6965
0.8385
0.2272
0.4505
1×10−3
0.9548
1.5478
0.8142
0.2121
0.4359
GNN layers
1
0.6946
1.0332
0.8082
0.2098
0.4258
2
1.0371
1.6965
0.8385
0.2272
0.4505
Appendix
Table 7: Sensitivity of DRL4TOD performance to the PPO learning rate and GNN message-passing depth at Xintiandi Station. Values are means over five independent runs. In each block, only the listed parameter is varied; all other settings remain fixed. Bold and underlining identify the reference configuration.
Model Variant
Performance Metrics
Total Reward
LUDE
LUDI
TCU
DF
Initial Plan
0.1757
0.1732
0.7858
0.1993
-
DRL4TOD
1.0371
1.6965
0.8385
0.2272
0.4505
Preservation-Constrained Scenario
0.6889
1.0002
0.7313
0.2598
0.4362
Appendix
Table 8: Performance comparison under heritage-preservation constraints at Xintiandi Station.
Figure 15: 3D visualization of the preservation-constrained scenario at Xintiandi Station: (a) initial planning configuration and (b) preservation-constrained configuration. Solid outlines in both panels indicate the parcels fixed in the preservation-constrained scenario.
Urban planning uses geospatial data, machine learning, and optimization to analyze cities and guide transport, land use, housing, and climate decisions.
We tackle the Metro Network Expansion Problem (MNEP), a subset of the Transport Network Design Problem (TNDP), which focuses on expanding metro systems to satisfy travel demand. Traditional methods rely on exact and heuristic approaches that require expert-defined constraints to reduce the search space. Recently, deep reinforcement learning (Deep RL) has emerged due to its effectiveness in complex sequential decision-making processes-it remains, however, computationally expensive, environmentally costly, and requires additional engineering to interpret. We show that MNEP problems are small enough to not require Deep RL methods. Reformulating the MNEP as a Non-Markovian Rewards Decision Process (NMRDP), we use tabular RL to achieve similar performance with significantly fewer training episodes, additionally offering greater interpretability. Additionally, we incorporate social equity criteria into the reward functions, focusing on efficiency and fairness, highlighting the versatility of our method. Evaluated in real-world settings-Xi'an and Amsterdam-our method reduces total episodes by a factor of 18 and total carbon emissions by a factor of 12 on average, while remaining competitive with Deep RL. This approach offers a replicable, modular, interpretable, and resource-efficient solution with potential applications to other combinatorial optimization problems.
Dimitris Michailidis, Sennay Ghebreab, Fernando P. Santos
Socially Intelligent Artificial Systems University of Amsterdam
Designing a transit network requires many sequential route extension decisions, but their quality is often visible only after the full network is assembled. This delayed-feedback challenge lies at the heart of the Transit Route Network Design Problem (TRNDP), where route interactions can be deceptive: an extension that appears useful locally can create transfer bottlenecks, produce redundant overlap, or reduce overall throughput. To guide route construction under delayed simulator feedback, we introduce AlphaTransit, a search-based planning framework for cityscale bus network design. AlphaTransit couples Monte Carlo Tree Search (MCTS) with a neural policy-value network: the policy proposes route extensions, the value estimates downstream design quality, and search uses these predictions to refine each decision. This provides decision-time lookahead during route construction without running simulator rollouts inside the search tree. We evaluate AlphaTransit on a new Bloomington TRNDP benchmark with realistic road topology and censusderived demand, under mixed and full transit demand settings. In the Bloomington network, AlphaTransit attains the highest service rate in both demand settings, reaching 54.6% and 82.1%, respectively. Relative to reinforcement learning without search, these correspond to 9.9% and 11.4% service rate gains; relative to MCTS without learned guidance, they correspond to 2.5% and 11.2% gains. These results suggest that coupling learned guidance with MCTS is more effective than using either approach alone for transit network design. Our code and data are publicly available in https://github.com/poudel-bibek/AlphaTransit.
Bibek Poudel, Sai Swaminathan, Weizi Li
Department of EECS, University of Tennessee, Knoxville, TN, USA · Department of CSE, University of California, Riverside, CA, USA
Urban transportation networks present complex optimization challenges spanning calibration of high-fidelity simulators and real-time operational control. This paper presents a shared latent-space framework that connects simulator calibration and reinforcement learning control through a common learned representation of urban traffic dynamics. First, we develop a combinatorial MLP-autoencoder architecture that learns low-dimensional manifolds linking simulator inputs (origin-destination demand, network parameters) to outputs (travel times, congestion patterns), enabling efficient Bayesian optimization for calibration. This approach demonstrates superior sample efficiency compared to traditional dimension reduction methods, achieving better fit to observational data within fixed computational budgets. Second, we implement a deep Q-learning agent with experience replay and target networks to optimize dynamic traffic assignment through scheduling and routing adjustments. In empirical evaluations on benchmark networks, our approach reduces system-wide travel times by up to 51% compared to baseline operations. The learned latent representation is not only used to reduce the dimensionality of Bayesian calibration, but is also incorporated into the reinforcement learning state representation, allowing the control policy to operate on compressed and calibrated traffic dynamics. This shared latent-space formulation provides a unified pathway from simulator calibration to adaptive operational control within intelligent transportation systems. Our results highlight the transformative potential of deep learning methods in urban mobility planning and management, particularly for large-scale networks where traditional optimization approaches face computational bottlenecks.
Adewumi Augustine Adepitan, Christopher J. Haruna, Oluwasegun Adegoke +2
Dept. of Civil, Environmental and Infrastructure Engineering George Mason University Fairfax, VA, USA · Dept. of Sustainability University of South Dakota Vermillion, SD, USA · School of Information Studies Syracuse University Syracuse, NY, USA +2