Driving · 2026-10-12
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents
Jie Huang, Yanan Chen, Ruixun Liu, Kaichen He, Xinyi He, Rui Huang, Zilong Zheng, Zhenliang Zhang, Yiwu Zhong
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
EmbodiedCLBench is introduced as “the first benchmark in embodied simulation” to evaluate continual learning for self-evolving agents. It uses compositional generalization: agents first learn basic embodied tasks, then are tested on unseen advanced tasks requiring recombination of learned skills. Experiments across multiple harnesses/models show advanced tasks are hard in zero-shot, but basic-task experience yields consistent improvement that stays stable with complexity yet does not scale with more learning samples.
Why it matters
The paper targets a gap: continual learning benchmarks are largely non-embodied, while embodied tasks add spatial/physical constraints. EmbodiedCLBench operationalizes continual learning via compositional generalization and provides an offline, deterministic evaluation protocol based on state-transition graphs. Reported findings characterize when experience helps (transfer to advanced tasks) and where it fails (lack of scaling with additional samples), offering guidance for evaluating and designing self-evolving embodied agents.
Method
- Introduce EmbodiedCLBench: basic tasks (built from TongSim atomic object–action interactions) and advanced tasks (composing elements across multiple basic-task categories) evaluated via compositional generalization.
- Use an offline, reproducible state-transition graph for each task: each node has a first-person image and candidate actions (optimal, executable detours, invalid), letting agents interact without online simulation.
- Evaluate three settings—Zero-shot, In-Context Learning (keep basic trajectories in context), and Skill Learning (provide a concise Markdown skill summary). Report success rate under step budget constraints across multiple agent harnesses and backbone models.
Limitation
While the step budgets are based on shortest-path length plus extra tolerance, the paper’s results highlight that “this improvement does not scale accordingly with the number of learning samples,” and that “current agents cannot extract meaningful information from additional experience.”
Abstract (from arXiv)
Continual learning requires models to accumulate past experience and transfer it to new tasks. Despite substantial research on continual learning, existing work remains largely confined to non-embodied settings, without extending to embodied environments. To address this gap, we introduce EmbodiedCLBench, the first benchmark in embodied simulation for evaluating the continual learning capability of self-evolving agents. With compositional generalization as the core evaluation principle, our benchmark evaluates whether agents can first learn from basic tasks and then recombine the acquired skills to solve complex, unseen advanced tasks. Based on EmbodiedCLBench, we conduct extensive experiments across multiple agent harnesses and learning methods, revealing consistent limitations and distinctive behaviors. Our results show that while advanced tasks can be hardly solved under zero-shot setting, the experience from basic tasks enables consistent improvement. Such benefit from the experience remains stable even if the task complexity increases. However, this improvement does not scale accordingly with the number of learning samples, and current agents cannot extract meaningful information from additional experience. Collectively, our findings highlight continual learning as a fundamental yet underexplored capability, and our benchmark offers valuable resources for designing effective self-evolving agents in embodied environments. Our code and dataset are available at https://pku-value-lab.github.io/EmbodiedCLBench-homepage.
Related papers
- EmbodiedRSI: Active Continual Robot Learning Through Hypothesis-Guided Co-Evolution
- RoboBridge: A Self-Evolving Embodied Agent Framework for Sim-to-Real Transfer
- SEAR: Self-Evolving Agentic Robotics
- Explore, Execute, Evolve: A Skill Acquisition and Reuse Loop for Embodied Agents
- AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-09 · Indexed by Pub-AI: 2026-10-12 UTC
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents
Preprint posted: 2026-10-09 · Original record ↗
- arxiv
- 2610.13021
Explicit source versions (1)
- v1 · 2026-10-09
- Preprint available
Source date: 2026-10-09 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Continual learning requires models to accumulate past experience and transfer it to new tasks. Despite substantial research on continual learning, existing work remains largely confined to non-embodied settings, without extending to embodied environments. To address this gap, we introduce EmbodiedCLBench, the first benchmark in embodied simulation for evaluating the continual learning capability of self-evolving agents. With compositional generalization as the core evaluation principle, our benchmark evaluates whether agents can first learn from basic tasks and then recombine the acquired skills to solve complex, unseen advanced tasks. Based on EmbodiedCLBench, we conduct extensive experiments across multiple agent harnesses and learning methods, revealing consistent limitations and distinctive behaviors. Our results show that while advanced tasks can be hardly solved under zero-shot setting, the experience from basic tasks enables consistent improvement. Such benefit from the experience remains stable even if the task complexity increases. However, this improvement does not scale accordingly with the number of learning samples, and current agents cannot extract meaningful information from additional experience. Collectively, our findings highlight continual learning as a fundamental yet underexplored capability, and our benchmark offers valuable resources for designing effective self-evolving agents in embodied environments. Our code and dataset are available at https://pku-value-lab.github.io/EmbodiedCLBench-homepage.