EmbodiedCLBench: First Benchmark for Continual Learning in Embodied Agents
PKU-VaLuE-Lab · hf · 2026-10-12
Peking University, BIGAI, and HKU introduce EmbodiedCLBench, the first embodied-simulation benchmark evaluating continual learning in self-evolving agents, built on compositional generalization: learn basic tasks, then recombine skills to solve complex unseen ones.
Findings:
- Advanced tasks are nearly unsolvable zero-shot, but experience from basic tasks yields consistent improvement that stays stable as complexity grows
- The benefit doesn't scale with the number of learning samples — current agents can't extract meaningful information from additional experience
Continual learning is framed as a fundamental yet underexplored capability; code and dataset are open-sourced.
More from Embodied
- Shenzhen robot map: 36 companies, most humanoid robot firms of any Chinese city — FinanceYF5 · 2026-10-12
- Hangzhou robot map: 25 companies led by Unitree, fastest productization in China — FinanceYF5 · 2026-10-12
- Shanghai robot map: 25 companies, industrial heavyweights FANUC, ABB, KUKA lead — FinanceYF5 · 2026-10-12
- Silicon Valley hosts 67 robot companies vs Shenzhen's 36: a 5-map comparison — FinanceYF5 · 2026-10-12
- Hive Robotics trains Unitree G1 to restock supermarket shelves, autonomous by 2027 — Distinct-Question-16 · 2026-10-12
- Mocap-suit running tracker goes from concept to product in one week — nick_linck · 2026-10-12