Zeva: In-Context Causal Learning for Generalizable Embodied Manipulation
cs.RO
Submitted: 2026-08-31
Updated: 2026-09-22
Project page: https://air-embodied-brain.github.io/Zeva
License: http://creativecommons.org/licenses/by/4.0/
The gist: Generalizable embodied manipulation remains difficult to achieve through pretraining alone, due to unseen physical conditions in the real world.
Terminology
Abstract
Generalizable embodied manipulation remains difficult to achieve through pretraining alone, due to unseen physical conditions in the real world. We argue that robots need to learn from their own physical interactions on the fly during real-world deployment and use this knowledge to inform subsequent actions. We present Zeva, the first framework that enables in-context learning from a robot's own physical interaction experience while keeping the policy model frozen. Zeva employs a Causal Interaction Extractor to encode an executed action and its induced state change into a causal interaction signal, which is stored in a dual-timescale causal memory. For subsequent actions, relevant causal interaction signals are retrieved from memory and injected into the frozen policy model as context. Experiments in simulation and real-world manipulation demonstrate that Zeva achieves the best performance among the compared frontier VLAs and WAMs and, more importantly, enables self-evolution during deployment without gradient updates. Its success rate continues to improve as the robot accumulates interaction experience. Furthermore, the acquired interaction experience can generalize across tasks.
Sources
- $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control
- RT-1: Robotics Transformer for Real-World Control at Scale
- Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning
- Causal World Modeling for Robot Control
- World Action Models are Zero-shot Policies
- RoboTTT: Context Scaling for Robot Policies
- WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time
- $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization
- $\pi^{*}_{0.6}$: a VLA That Learns From Experience
- GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulation
- FAST: Efficient Action Tokenization for Vision-Language-Action Models
- PACT: Perception-Action Causal Transformer for Autoregressive Robotics Pre-Training
- Action-Effect Memory Pretraining for Robot Manipulation
- ALAM: Algebraically Consistent Latent Action Model for Vision-Language-Action Models
- UniVLA: Learning to Act Anywhere with Task-centric Latent Actions
- MEM: Multi-Scale Embodied Memory for Vision Language Action Models
- DIM-WAM: World-Action Modeling with Diverse Historical Event Memory
- MemoryWAM: Efficient World Action Modeling with Persistent Memory
- VLA-RL: Towards Masterful and General Robotic Manipulation with Scalable Reinforcement Learning
- Fast-WAM: Do World Action Models Need Test-time Future Imagination?
Related papers
- FMT x: An Efficient and Asymptotically Optimal Extension of the Fast Marching Tree for Dynamic Replanning
- MPCFormer: A physics-informed data-driven approach for explainable socially-aware autonomous driving
- RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies
- ARC: A Reasoning Recipe for Robot Foundation Models
- HRDexDB: A 4D Dexterous Grasping Dataset Across Human and Multiple Robot Embodiments
- APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies