The paper argues that zero‑shot generalization in robotics is limited by a lack of diverse training data, unlike digital AI where crowd‑sourced data drives progress. It defines zero‑shot generalization as coverage of the deployment distribution and shows that adding more samples from the same source only marginally reduces the unseen region, whereas new contributors significantly expand it. The authors propose a shared latent space for robot actions, combining multi‑modal alignment, generative augmentation, and noisy‑label learning, and suggest building a community of developers around a single standard robot, Origami, to generate diverse data for testing zero‑shot success.
General-purpose robots must handle tasks, objects and environments they were never trained on, yet zero-shot generalization remains one of the least solved problems in physical AI. In digital AI, zero...
arXiv:2507. 06219v2 Announce Type: replace-cross Abstract: Data scaling has driven remarkable success in foundation models for Natural Language Processing (NLP) and Computer Vision (CV), yet the principles of effective data scaling in robotic manipulation remain insufficiently understood.
arXiv:2607. 15880v1 Announce Type: cross Abstract: Imitation Learning aims to learn skills from extensive observations and demonstrations for robots, so it suffers from data scarcity and environment generalization.
HumanEgo is a framework that enables zero‑shot robot learning from short egocentric human videos by converting each demonstration into an entity‑level hand‑object interaction representation and training a flow‑matching policy with dense auxiliary objectives. The method is robot‑data‑free, hardware‑agnostic, and data‑efficient, achieving 92.5 % success on four real‑world tasks with only 30 minutes of human video per task and outperforming matched‑time robot teleoperation by 41 %. HumanEgo also robustly transfers zero‑shot across new robots, cameras, and environments, and is released as an open‑source tool for learning robot policies directly from human data.
Video Prediction Policy 2 (VPP2) is a new world action model that improves zero‑shot generalization for both video prediction and action generation. It achieves this by pretraining a large, diverse manipulation video dataset with event‑level supervision, then distilling the model into a single‑step visual planner and adding a mixture‑of‑transformers action module. Experiments show VPP2 outperforms leading baselines on open‑ended video prediction, real‑world zero‑shot manipulation, and several challenging benchmarks.
arXiv:2606. 10918v1 Announce Type: cross Abstract: The recent trend in scaling models for robot learning has resulted in impressive policies that can perform various manipulation tasks and generalize to novel scenarios.
Zero-WAM introduces a causal video-action model that enables robots to perform unseen manipulation tasks by following in-context human video guidance. The authors create HumanGen, a dataset of 74.2K human-robot ICL pairs across 8.6K tasks, and propose an in-context future chunk prediction objective to prevent shortcut learning. In simulation, Zero-WAM attains a 47.0% success rate on seven unseen tasks, outperforming the best video-action baseline by 29.5 percentage points, and demonstrates real‑world generalization to complex, long‑horizon, and fine‑grained tasks.
Zero-shot cross-task generalization, where a policy must execute manipulation tasks never seen during training, remains a central challenge in robot learning. In large language models, a novel task ca...
The paper introduces CASTER, a framework that formalizes the conversion of synthetic videos into real robot trajectories as a constrained optimization problem. It shows that synthetic videos must probabilistically seed the search within a semantic basin of attraction, allowing local trajectory refinement while maintaining task alignment. By removing simplifying assumptions from existing methods, the authors achieve more robust zero‑shot manipulation and improve task success by 34.1% over the best baseline.
The paper investigates zero‑shot task generalisation in offline multi‑agent reinforcement learning by extending sequence‑modeling architectures to support multi‑task observation and action spaces and variable agent counts. It finds that increasing task diversity, rather than merely enlarging the dataset, is the key driver for robust zero‑shot transfer. Experiments on four challenging environments show a 3.2× mean improvement on held‑out tasks compared to single‑task models and outperform strong behaviour‑cloning baselines.
ZimaBlue is a scalable framework that learns generalizable World Action Models (WAMs) from large-scale egocentric videos. It follows a three-stage curriculum: causal video pre‑training, video‑action mid‑training with a unified action representation, and final specialization to a target robot. The system employs an asynchronous Slow‑Fast architecture to enable real‑time 30 Hz action prediction, achieving a jump in real‑robot zero‑shot success from 36.1% to 77.8% when leveraging over 120,000 hours of embodied video.
The paper introduces QDTraj, a method that uses Quality‑Diversity algorithms to automatically generate a diverse set of low‑level trajectory primitives for manipulating articulated objects. By leveraging sparse reward exploration, QDTraj produces at least five times more diverse trajectories for hinge and slider tasks compared to baseline methods, and demonstrates strong generalization across 30 articulations from the PartNetMobility dataset, averaging 704 trajectories per task. The resulting primitives are validated both in simulation and on real robots, with the code released publicly.