Mobile Agent Navigation on AndroidControl 1.0 (test)
72.9Step Success Rate (SR)ADPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ADPONumber of samples (N)=122026.01 | 72.9 | 86.3 | 78.9 | |
| ADPONumber of samples (N)=42026.01 | 72.7 | 86.3 | 79.5 | |
| ADPONumber of samples (N)=82026.01 | 72.7 | 86.4 | 78.7 | |
| UI-TARS-7BNumber of samples (N)=12026.01 | 72.5 | 83.7 | — | |
| GRPONumber of samples (N)=122026.01 | 71.1 | 85.6 | 77.4 | |
| GRPONumber of samples (N)=12026.01 | 71 | 86 | 76.9 | |
| GRPONumber of samples (N)=42026.01 | 71 | 85.5 | 77.2 | |
| ADPONumber of samples (N)=12026.01 | 70.9 | 85.8 | 76.2 | |
| GRPONumber of samples (N)=82026.01 | 70.8 | 85.6 | 76.9 | |
| AgentCPM-GUI-8BNumber of samples (N)=12026.01 | 69.2 | 77.7 | — | |
| SpiritSight-8BNumber of samples (N)=12026.01 | 68.1 | — | — | |
| Qwen2.5-VL-7B (Base)Number of samples (N)=12026.01 | 61.3 | 82.2 | 73.6 | |
| Qwen2.5-VL-7B (Base)Number of samples (N)=82026.01 | 58.3 | 78.7 | 68.8 | |
| Qwen2.5-VL-7B (Base)Number of samples (N)=122026.01 | 58.3 | 78.9 | 68.7 | |
| Qwen2.5-VL-7B (Base)Number of samples (N)=42026.01 | 56 | 76.3 | 68.1 |