Preemptive Instruction Following with Resumption

Zerun Liu; Lechao Cheng; Yimin Liu; Shengeng Tang; Yaxiong Wang; Zhun Zhong

cs.ROOct 26, 2026

Preemptive Instruction Following with Resumption

Authors: Zerun Liu, Lechao Cheng, Yimin Liu, Shengeng Tang, Yaxiong Wang, Zhun Zhong

Organizations: School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, Anhui 230601, China

Abstract

Long-horizon embodied navigation may be interrupted by a higher-priority instruction before the current task sequence is complete. The agent must finish its current atomic task, execute the inserted task, and resume the remaining sequence from the updated environment state, where its viewpoint and available navigation candidates may have changed. To evaluate this capability, we introduce PIFR benchmark, a novel task for runtime task preemption and resumption in embodied navigation. PIFR includes single, repeated, and nested interruption settings,with stage-specific metrics for inserted-task completion, resumption of the original task, and completion of the full sequence. Evaluations show that recovery performance varies across interruption conditions, while aggregate success can obscure differences in interruption handling and resumption. We further propose a lightweight candidate-ranking baseline to improve target re-grounding after interruption. Together, PIFR and the baseline provide a systematic framework for studying interruption handling and task recovery in long-horizon embodied navigation.

Figures & tables

Figure 1: Interruption-and-resumption navigation in PIFR. During long-horizon navigation, an agent may need to suspend its current task, handle an inserted objective, and then return to the unfinished task. Blue, red, and green trajectories indicate base, interrupt, and resume execution, respectively.

Figure 2: The overview of the composition of PIFR and the Single, Repeated and Nested interruption. Among the three different types of interrupts—Single, Repeated, and Nested—the orange arrow represents an interrupt, and the blue arrow represents a resume.

Table 3

Figure 3: Overview of PIFR-Rank. The upper shows object and frontier candidate generation from visual observations and the active instruction. The lower illustrates the ranking objective, which encourages target-consistent candidates to score above distractors.

Method	❶ Single Interruption				❷ Repeated Interruption				❸ Nested Interruption
Method	SR $\uparrow$	SPL $\uparrow$	Int. $\uparrow$	Res. $\uparrow$	SR $\uparrow$	SPL $\uparrow$	Int. $\uparrow$	Res. $\uparrow$	SR $\uparrow$	SPL $\uparrow$	Int. $\uparrow$	Res. $\uparrow$
PIFR-SG3D
R2F	4.8	2.8	3.2	3.0	3.7	1.9	3.5	5.0	3.3	2.3	3.1	2.8
E-VA	4.7	3.5	3.2	6.3	4.9	3.4	2.5	10.0	6.5	4.7	5.7	10.4
Uni-NaVid	10.7	7.1	9.0	9.0	7.0	4.6	7.4	7.5	6.0	3.6	4.6	3.8
MTU3D	22.6	12.8	22.0	23.0	15.8	11.4	11.3	20.0	15.6	9.2	11.3	15.1

Table 3: Results on PIFR-SG3D / PIFR-OVON (Seen) / PIFR-GOAT (Seen) under different interruption settings ( ❶ Single Interruption, ❷ Repeated Interruption, ❸ Nested Interruption ).

Figure 4: Radar plots of SR, SPL, Interrupt SR, and Resume SR on the Synonym and Unseen splits of PIFR-OVON and PIFR-GOAT under single, repeated, and nested interruption.

Figure 5: Repeated and nested execution reveal distinct stage-wise recovery profiles. Resume SR is decomposed into the first and second resumptions for ABACA (

B\!\rightarrow\!A

and

C\!\rightarrow\!A

), and the inner and outer resumptions for ABCBA (

C\!\rightarrow\!B

and

B\!\rightarrow\!A

). Recovery behavior differs substantially across navigation settings, showing that a single ABA recovery score cannot fully characterize repeated and nested task resumption.

Method	SR $\uparrow$	SPL $\uparrow$	Int. SR $\uparrow$	Res. SR $\uparrow$
Baseline	37.03	21.55	31.34	40.82
Baseline + BCE.	39.82	23.32	33.46	44.91
Baseline + BCE. + CR.	39.28	23.54	34.36	43.10
Baseline + BCE. + CR. + RW.	39.21	23.25	35.25	45.04

Table 4: Ablation study of PIFR-Rank. BCE. , CR. and RW. denote BCE Loss, candidate ranking and resumption weighting, respectively. The best result for each metric is highlighted in bold.

Embodied Navigation

Embodied navigation studies how robots perceive, map, and move through environments to reach goals and follow instructions.

See all 138 papers

When Users Change Their Mind: Evaluating Interruptible Agents in Long-Horizon Web Navigation
Apr 1, 2026Henry Peng Zou, Chunyu Miao, Wei-Chieh Huang +16LLM Agents LLM Agent Evaluation
University of Illinois Chicago · McGill University · MBZUAI +2
NavHarness: Towards Lifelong Embodied Navigation
Sep 28, 2026Xunyi Zhao, Jian Zhou, Sihao Lin +6Embodied Navigation Memory-Augmented Agents
Adelaide University · Responsible AI Research Centre, AIML · CSIRO Data61
Lifelong Embodied Navigation Learning
Mar 6, 2026Xudong Wang, Jiahua Dong, Baichen Liu +3Embodied Navigation Continual Learning for LLM Agents
State Key Laboratory of Robotics and Intelligent Systems, Shenyang Institute of Automation, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Mohamed bin Zayed University of Artificial Intelligence

Apr 1, 2026J/K move · Enter open · S save

When Users Change Their Mind: Evaluating Interruptible Agents in Long-Horizon Web Navigation

Henry Peng Zou, Chunyu Miao, Wei-Chieh Huang, Yankai Chen+15

University of Illinois Chicago · McGill University · MBZUAI+2

LLM Agents LLM Agent Evaluation Web Agents Web Agent Benchmarks Long-Horizon Agent Evaluation

Open paper AskPDF arXiv

As LLM agents transition from short, static problem solving to executing complex, long-horizon tasks in dynamic environments, the ability to handle user interruptions, such as adding requirement or revising goals, during mid-task execution is becoming a core requirement for realistic deployment. However, existing benchmarks largely assume uninterrupted agent behavior or study interruptions only in short, unconstrained language tasks. In this paper, we present the first systematic study of interruptible agents in long-horizon, environmentally grounded web navigation tasks, where actions induce persistent state changes. We formalize three realistic interruption types, including addition, revision, and retraction, and introduce InterruptBench, a benchmark derived from WebArena-Lite that synthesizes high-quality interruption scenarios under strict semantic constraints. Using a unified interruption simulation framework, we evaluate six strong LLM backbones across single- and multi-turn interruption settings, analyzing both their effectiveness in adapting to updated intents and their efficiency in recovering from mid-task changes. Our results show that handling user interruptions effectively and efficiently during long-horizon agentic tasks remains challenging for powerful large-scale LLMs. Code and dataset are available at https://github.com/HenryPengZou/InterruptBench.

Abstract

Embodied Navigation

Explore similar work

When Users Change Their Mind: Evaluating Interruptible Agents in Long-Horizon Web Navigation

NavHarness: Towards Lifelong Embodied Navigation

Lifelong Embodied Navigation Learning