AndroTMem: From Interaction Trajectories to Anchored Memory in Long-Horizon GUI Agents
Yibo Shi 1,∗ , Jungang Li 2,3,∗,† , Linghao Zhang 4,∗ , Zihao Dongfang 2,3,∗ , Biao Wu 5 , Sicheng Tao 2 , Yibo Yan 2,3 , Chenxi Qin 6 , Weiting Liu 7 , Zhixin Lin 8 , Hanqian Li 2 , Yu Huang 2 , Song Dai 2,3 , Yonghua Hei 2,3 , Yue Ding 9 , Xiang Li 2,3 , Shikang Wang 4 , Chengdong Xu 10 , Jingqi Liu 1 , Xueying Ma 1 , Zhiwen Zheng 1 , Xiaofei Zhang 1 , Bincheng Wang 11 , Nichen Yang 1 , Jie Wu 10 , Lihua Tian 1,‡ , Chen Li 1,‡ , Xuming Hu 2,3,‡ 1 XJTU 2 HKUST(GZ) 3 HKUST 4 CityU 5 UTS 6 TJU 7 FDU 8 SDU 9 CASIA 10 SYSU 11 NWPU ∗ Core Contributors. † Project Lead. ‡ Corresponding Authors. Contact: [email protected] , [email protected] , [email protected] , [email protected] · 2026
Read AndroTMem: From Interaction Trajectories to Anchored Memory in Long-Horizon GUI Agents by Yibo Shi 1,∗ , Jungang Li 2,3,∗,† , Linghao Zhang 4,∗ , Zihao Dongfang 2,3,∗ , Biao Wu 5 , Sicheng Tao 2 , Yibo Yan 2,3 , Chenxi Qin 6 , Weiting Liu 7 , Zhixin Lin 8 , Hanqian Li 2 , Yu Huang 2 , Song Dai 2,3 , Yonghua Hei 2,3 , Yue Ding 9 , Xiang Li 2,3 , Shikang Wang 4 , Chengdong Xu 10 , Jingqi Liu 1 , Xueying Ma 1 , Zhiwen Zheng 1 , Xiaofei Zhang 1 , Bincheng Wang 11 , Nichen Yang 1 , Jie Wu 10 , Lihua Tian 1,‡ , Chen Li 1,‡ , Xuming Hu 2,3,‡ 1 XJTU 2 HKUST(GZ) 3 HKUST 4 CityU 5 UTS 6 TJU 7 FDU 8 SDU 9 CASIA 10 SYSU 11 NWPU ∗ Core Contributors. † Project Lead. ‡ Corresponding Authors. Contact: [email protected] , [email protected] , [email protected] , [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-18429.txt.
Description
Shi and co-authors present AndroTMem, a diagnostic framework for anchored memory in long-horizon Android GUI agents. Full replay is redundant and noisy; summaries erase key dependency details. Their bench holds 1,069 tasks with 34,473 steps built on strong step-to-step causal links. Longer tasks fail mostly from memory lapses, not perception or single action errors, they find. Anchored State Memory stores linked intermediate states and lifts task completion by 5 to 30%.
Catalog record
| CREATOR | Yibo Shi 1,∗ , Jungang Li 2,3,∗,† , Linghao Zhang 4,∗ , Zihao Dongfang 2,3,∗ , Biao Wu 5 , Sicheng Tao 2 , Yibo Yan 2,3 , Chenxi Qin 6 , Weiting Liu 7 , Zhixin Lin 8 , Hanqian Li 2 , Yu Huang 2 , Song Dai 2,3 , Yonghua Hei 2,3 , Yue Ding 9 , Xiang Li 2,3 , Shikang Wang 4 , Chengdong Xu 10 , Jingqi Liu 1 , Xueying Ma 1 , Zhiwen Zheng 1 , Xiaofei Zhang 1 , Bincheng Wang 11 , Nichen Yang 1 , Jie Wu 10 , Lihua Tian 1,‡ , Chen Li 1,‡ , Xuming Hu 2,3,‡ 1 XJTU 2 HKUST(GZ) 3 HKUST 4 CityU 5 UTS 6 TJU 7 FDU 8 SDU 9 CASIA 10 SYSU 11 NWPU ∗ Core Contributors. † Project Lead. ‡ Corresponding Authors. Contact: [email protected] , [email protected] , [email protected] , [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 76515 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.18429