ActivityEditor: Learning to Synthesize Physically Valid Human Mobility

Chenjie Yang; Yutian Jiang; Anqi Liang; Wei Qi; Chenyu Wu; Junbo Zhang

cs.AIApr 7, 2026

ActivityEditor: Learning to Synthesize Physically Valid Human Mobility

Authors: Chenjie Yang, Yutian Jiang, Anqi Liang, Wei Qi, Chenyu Wu, Junbo Zhang

Organizations: Southwest Jiaotong University · HKUST (Guangzhou) · Shanghai Jiao Tong University · Tsinghua University ·

Abstract

Human mobility modeling is indispensable for diverse urban applications. However, existing data-driven methods often suffer from data scarcity, limiting their applicability in regions where historical trajectories are unavailable or restricted. To bridge this gap, we propose ActivityEditor, a novel dual-LLM-agent framework designed for zero-shot cross-regional trajectory generation. Our framework decomposes the complex synthesis task into two collaborative stages. Specifically, an intention-based agent, which leverages demographic-driven priors to generate structured human intentions and coarse activity chains to ensure high-level socio-semantic coherence. These outputs are then refined by editor agent to obtain mobility trajectories through iteratively revisions that enforces human mobility law. This capability is acquired through reinforcement learning with multiple rewards grounded in real-world physical constraints, allowing the agent to internalize mobility regularities and ensure high-fidelity trajectory generation. Extensive experiments demonstrate that ActivityEditor achieves superior zero-shot performance when transferred across diverse urban contexts. It maintains high statistical fidelity and physical validity, providing a robust and highly generalizable solution for mobility simulation in data-scarce scenarios. Our code is available at: https://anonymous.4open.science/r/ActivityEditor-066B.

Figures & tables

Figure 1: Overview of the ActivityEditor framework.

Traditional Baselines
Method	California			Arizona			Georgia			Oklahoma
Method	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$
$\diamond$ DeepMove	0.554	0.660	0.229	0.521	0.666	0.252	0.448	0.657	0.264	0.472	0.676	0.281
$\diamond$ LSTPM	0.402	0.705	0.290	0.420	0.691	0.300	0.379	0.693	0.288	0.376	0.697	0.287
General LLM Baselines
$\diamond$ ChatGPT	0.660	0.425	0.199	0.664	0.399	0.195	0.583	0.403	0.201	0.616	0.404	0.158

Table 1: Performance comparison across four states under different models and settings.

Model	CA $\rightarrow$ CA			CA $\rightarrow$ AZ			CA $\rightarrow$ GA			CA $\rightarrow$ OK			AZ $\rightarrow$ CA
Model	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$	Acc $\uparrow$	Mint $\downarrow$	Atype $\downarrow$
SFT_CoPB	0.743	0.289	0.154	0.682	0.420	0.127	0.642	0.303	0.153	0.711	0.325	0.114	0.654	0.371	0.109
SFT_RAGHome	0.691	0.349	0.176	0.671	0.322	0.183	0.643	0.324	0.134	0.672	0.246	0.120	0.636	0.287	0.151
SFT_LLaMA	0.724	0.214	0.110	0.689	0.256	0.158	0.632	0.253	0.195	0.650	0.227	0.165	0.653	0.228	0.141
SFT_Qwen	0.753	0.231	0.104	0.745	0.239	0.113	0.708	0.248	0.130	0.714	0.241	0.127	0.699	0.214	0.097
GRPO_LLaMA	0.732	0.248	0.122	0.720	0.260	0.131	0.681	0.276	0.147	0.690	0.268	0.142	0.671	0.233	0.118

Table 2: Cross-state transfer performance. Models (SFT_LLaMA, SFT_Qwen, GRPO_LLaMA, GRPO_Qwen ) are implemented on the multi-agent architecture. CA, AZ, GA, and OK denote California, Arizona, Georgia, and Oklahoma, respectively.

Category	Metrics
Sequence Similarity	accuracy ( Acc ), f1_score, edit_dist, bleu_score
Temporal Alignment	macro_hour, micro_hour, micro_int, macro_int ( Mint )
Distributional Consistency	data_jsd, act_type ( Atype ), uni_act_type, traj_len

Table 3: Evaluation metrics.

Panel of Figure 2 — Figure 2: Ablation studies on the proposed ActivityEditor.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 3: Activity number distribution by state.

Figure 4: Activity type distribution by state.

Figure 6: Chain length distribution by state.

Hyperparameter	SFT Stage
Base Model	Qwen3-8B
LoRA Rank (r)	32
LoRA Alpha	64
LoRA Dropout	0.05
Training Epochs	3 epochs
Peak Learning Rate	5e-5

Appendix

Table 4: Hyperparameter settings for the SFT stage.

Hyperparameter	GRPO Stage
Base Model	Qwen3-8B
LoRA Rank (r)	32
LoRA Alpha	64
LoRA Dropout	0.05
Training Steps	1000 steps
Peak Learning Rate	1e-5

Appendix

Table 5: Hyperparameter settings for the GRPO stage.

Initial Draft	Reasoning Cues	Edited Output	Ground Truth
[00:00–07:45] Home		[00:00–07:45] Home	[00:00–07:45] Home
[07:45–16:45] Work	occupation $\rightarrow$ Evening Work	[07:45–16:30] Work	[07:45–16:30] Work
[16:45–18:00] Home	relationship $\rightarrow$ Shopping	[16:30–17:20] Shopping	[16:30–17:30] Shopping
[17:45–19:00] Work	constraint $\rightarrow$ Fix overlap	[17:20–19:50] Work	[17:30–17:45] Service
[19:00–19:35] Shopping	short commute $\rightarrow$ no extra Home stay	[19:50–24:00] Home	[17:45–20:30] Work
[19:35–24:00] Home			[20:30–24:00] Home

Appendix

Table 6: Case study of profile-grounded trajectory editing.

Trajectory Generation

Trajectory generation creates plausible, goal-directed paths and motion sequences for people, vehicles, and robots.

See all 182 papers

TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation
Jun 10, 2026Siyu Li, Toan Tran, Lingyi Zhao +2Trajectory Generation Human Behavior Simulation
Dept. of Computer Science, Emory University, Atlanta, GA, USA · Novateur Research Solutions, Ashburn, VA, USA
ELLMob: Event-Driven Human Mobility Generation with Self-Aligned LLM Framework
Mar 9, 2026Yusong Wang, Chuang Yang, Jiawei Wang +5
Institute of Science Tokyo · The University of Tokyo · The University of Osaka
Plan-to-Synthesis: Cross-City Human Mobility Generation via Semantic Latent Flow Matching
Sep 26, 2026Zhoufu Wang, Baoshen Guo, Zhiqing Hong +6Trajectory Generation Generative Modeling
Singapore-MIT Alliance for Research and Technology Centre, MIT · Nanyang Technological University · Hong Kong University of Science and Technology (Guangzhou) +3

Jun 10, 2026J/K move · Enter open · S save

TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation

Siyu Li, Toan Tran, Lingyi Zhao, Khurram Shafique+1

Dept. of Computer Science, Emory University, Atlanta, GA, USA · Novateur Research Solutions, Ashburn, VA, USA

Trajectory Generation Human Behavior Simulation Synthetic Data Generation LLM Agent Trajectory Synthesis LLM Agent Planning

Open paper AskPDF arXiv

Human mobility data is important for transportation, urban planning, and epidemic control, but large-scale trajectory collection is often costly and privacy-constrained, motivating realistic synthetic trajectory generation. Existing LLM-based generators typically rely on either prompt engineering, which preserves zero-shot reasoning but lacks fine-grained spatiotemporal grounding, or trajectory-level fine-tuning, which improves statistical precision but incurs substantial computational cost and may weaken general reasoning. We propose TrajGenAgent, a semantic-aware hierarchical LLM-agent framework for human mobility trajectory generation without model fine-tuning. TrajGenAgent uses a two-stage orchestrator-worker design: an LLM first synthesizes an individual- and weekday-conditioned activity chain from historical evidence via in-context learning, and a deterministic workflow then grounds each activity into a complete visit using personalized POI retrieval, distance-aware location selection, kinematics-aware travel-time propagation, and LLM-based duration estimation. To evaluate realism beyond aggregate spatiotemporal statistics, we introduce an anomaly-detection-based evaluation framework using two complementary detectors to assess behavioral and semantic plausibility. Experiments on benchmark and large-scale simulation datasets show that TrajGenAgent improves spatiotemporal fidelity, semantic coherence, and individual-specific behavioral realism over representative neural and LLM-based baselines, while avoiding parameter updates.

Abstract

Trajectory Generation

Explore similar work

TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation

ELLMob: Event-Driven Human Mobility Generation with Self-Aligned LLM Framework

Plan-to-Synthesis: Cross-City Human Mobility Generation via Semantic Latent Flow Matching