Organizations: The State Key Laboratory of Complex and Critical Software Environment, Beihang University China · H3I, Beihang University · China Mobile Information Technology Center China
Mobile devices continuously interact with cellular base stations, generating massive volumes of signaling records that provide broad coverage for understanding human mobility. However, such records offer only coarse location cues (e.g., serving-cell identifiers) and therefore limit their direct use in applications that require high-precision GPS trajectories. This paper studies the Sig2GPS problem: reconstructing GPS trajectories from cellular signaling. Inspired by domain experts often lay the signaling trace on the map and sketch the corresponding GPS route, unlike conventional solutions that rely on complex multi-stage engineering pipelines or regress coordinates, Sig2GPS is reframed as an image-to-video generation task that directly operates in the map-visual domain: signaling traces are rendered on a map, and a video generation model is trained to draw a continuous GPS path. To support this paradigm, a paired signaling-to-trajectory video dataset is constructed to fine-tune an open-source video model, and a trajectory-aware reinforcement learning-based optimization method is introduced to improve generation fidelity via rewards. Experiments on large-scale real-world datasets show substantial improvements over strong engineered and learning-based baselines, while additional results on next GPS prediction indicate scalability and cross-city transferability. Overall, these results suggest that map-visual video generation provides a practical interface for trajectory data mining by enabling direct generation and refinement of continuous paths under map constraints.
Figures & tables
Figure 1. Overview of the Sig2GPS problem: coarse cellular signaling records are mapped to high-precision GPS trajectories via video generation.
Figure 2. Illustration of Think Over Trajectory. We highlight the difference between previous seq2seq methods, VLM-based methods and our proposed methods.
Figure 3. Overall architecture of our Think Over Trajectory framework for Sig2GPS. We first fine-tune the video generator on paired signaling-trajectory videos, and then apply trajectory-aware reinforcement learning from verifiable reward (RLVR) to further improve fidelity under map and temporal constraints.
Figure 4. Three Examples of Our Roll Out.
Small
Medium
Large
Model
MAE ↓
RMSE ↓
L100 ↑
G1000 ↓
MAE ↓
RMSE ↓
L100 ↑
G1000 ↓
MAE ↓
RMSE ↓
L100 ↑
G1000 ↓
GRU
402.52
475.11
5.18%
2.81%
489.37
626.72
5.24%
7.35%
647.24
829.69
2.37%
15.06%
GPT
358.14
428.77
7.04%
2.07%
473.56
636.63
6.72%
6.63%
655.60
842.04
3.89%
16.59%
MLP
513.62
635.87
4.67%
10.57%
648.33
846.51
4.28%
18.13%
827.79
1016.15
1.26%
28.68%
TCN
613.97
737.26
2.63%
13.53%
722.73
873.25
1.60%
20.30%
951.42
1138.49
0.90%
36.71%
TrajFormer
375.12
431.60
5.12%
1.92%
402.13
485.89
6.20%
4.31%
489.70
579.33
3.62%
15.72%
Table 1. Performance comparison. Lower is better for MAE/RMSE/G1000; higher is better for L100.
Frame Count
Small
Medium
Large
MAE ↓
RMSE ↓
MAE ↓
RMSE ↓
MAE ↓
RMSE ↓
13
239.64
323.69
323.83
428.35
600.14
919.35
17
230.20
318.49
318.49
440.20
546.76
801.46
21
214.96
268.26
307.85
367.31
441.10
761.93
Table 2. Test-time scaling behavior with different numbers of input frames. Lower is better for both MAE and RMSE.
Method
Chengdu
Xi’an
MAE ↓
RMSE ↓
MAE ↓
RMSE ↓
DeepMove
285.36
421.41
486.52
562.46
GetNext
239.43
316.35
577.96
661.53
UniMob
239.43
354.40
416.56
485.47
AgentMove
221.15
323.90
428.81
498.34
STAN
257.19
377.90
455.15
523.71
Table 3. Performance Comparison on Next GPS Prediction.
Figure 5. Ablation results on Sig2GPS.
Figure 6. Effect of KL regularization strategies.
Figure 7. Representative case of Sig2GPS video generation.
Figure 8. System overview of the deployed Sig2GPS service.
The importance of mobile phone GPS trajectory data is widely recognized across many fields, yet the use of real data is often hindered by privacy concerns, limited accessibility, and high acquisition costs. As a result, generating pseudo-GPS trajectory data has become an active area of research. Recent diffusion-based approaches have achieved strong fidelity but remain limited in spatial scale (small urban areas), transportation-mode diversity, and efficiency (requiring numerous sampling steps). To address these challenges, we introduce TrajFlow, which to the best of our knowledge is the first flow-matching-based generative model for GPS trajectory generation. TrajFlow leverages the flow-matching paradigm to improve robustness and efficiency across multiple geospatial scales, and incorporates a trajectory harmonization and reconstruction strategy to jointly address scalability, diversity, and efficiency. Using a nationwide mobile phone GPS dataset with millions of trajectories across Japan, we show that TrajFlow or its variants consistently outperform diffusion-based and deep generative baselines at urban, metropolitan, and nationwide levels. As the first nationwide, multi-scale GPS trajectory generation model, TrajFlow demonstrates strong potential to support inter-region urban planning, traffic management, and disaster response, thereby advancing the resilience and intelligence of future mobility systems.
Peiran Li, Jiawei Wang, Haoran Zhang +4
Hitotsubashi University · The University of Tokyo · LocationMind Inc. +1
Urban trajectories play a crucial role in modeling urban dynamics and supporting various smart city applications. However, privacy concerns restrict access to large-scale and high-quality trajectory datasets. Trajectory generation provides a promising alternative by synthesizing realistic data to mitigate privacy risks. However, existing methods fail to explicitly capture travel patterns and can only generate fixed-length trajectories under a single condition. To address these limitations, we propose HTP, which Hierarchically generates Travel patterns first and then generates GPS Points by using large language models (LLMs), rather than directly generating GPS points. We first design a trajectory-specific residual quantization variational autoencoder (RQ-VAE) that quantizes micro-level GPS trajectories into compact, macro-level travel pattern tokens in a coarse-to-fine manner. These tokens capture rich segment spatial irregularities, such as point density variations caused by traffic conditions. Then, we extend the LLM vocabulary with travel pattern tokens to align trajectory representations with the LLM input, and apply supervised fine-tuning (SFT) to align the LLM with the trajectory generation task, enabling generation of travel pattern sequences under various conditions. Extensive experiments on two real-world datasets show that HTP outperforms the strongest baseline by an average of 29.78% in terms of generation quality. Our code is available at https://github.com/slzhou-xy/HTP.
Silin Zhou, Chenhao Wang, Yuntao Wen +3
University of Electronic Science and Technology of China Chengdu, China · King Abdullah University of Science and Technology Thuwal, Saudi Arabia
Geospatial simulation of infrastructure interventions requires mobility generators that respond directly to edited maps, yet many data-driven generators do not expose the map as an editable condition. We formulate this task as map-conditioned autoregressive generation of human mobility: a road raster conditions a decoder that emits nominal 31.25 m mesh-cell tokens at one-minute intervals. The mesh-local vocabulary supports held-out and locally edited maps without retraining or vocabulary changes. We instantiate a ResNet-50 visual-prefix configuration and a Vision Transformer (ViT) cross-attention configuration, trained from scratch on 87,400 smartphone-derived trajectories from 874 meshes in Ishikawa Prefecture, Japan; 219 meshes are held out. We evaluate map sensitivity by comparing correct-map and within-split shuffled-map generations with held-out real trajectories. On the 110-mesh test split, for the ResNet-50 configuration, correct-map generations are closer than shuffled-map generations on 60% of meshes under Hausdorff-based energy distance (p = 0.021), while DTW is directional but inconclusive (57%, p = 0.074); correlation with real density is 0.38 with the correct map versus 0.01 with shuffled maps. The ViT configuration shows weaker trajectory-level sensitivity and smaller density gains. An illustrative bridge-removal edit changes generated continuations without retraining. Together, these results support the feasibility of editable-map human-mobility simulation.