Building Scalable Infrastructure for Robot Learning:
From Data Scaling to Real-World Deployment

This workshop focuses on four interconnected pillars:

• Data Scaling: Exploring scalable strategies for acquiring, generating, and orchestrating multimodal data across real-world interactions and simulations.
• Training and Post-Training: Examining supervised fine-tuning, imitation learning, reinforcement learning, preference learning, and scalable training workflows for embodied agents.
• Evaluation: Developing standardized benchmarking protocols and methodologies to assess generalization, long-horizon reasoning, safety, robustness, and data efficiency.
• Deployment Infrastructure: Building reliable software-hardware integration pipelines and continuous feedback loops to support real-world robot execution and fleet learning.

Scope of Accepted Papers

We invite submissions that address the system-level challenges in scalable robot learning. Topics of interest include, but are not limited to:

• Real-world robot interaction data collection and large-scale teleoperation pipelines
• Simulation data generation, synthetic multimodal data, and sim-to-real transfer
• Supervised fine-tuning, imitation learning, and post-training workflows for VLA models
• Offline and online reinforcement learning, and preference learning for embodied agents
• Benchmarking VLA capabilities, long-horizon reasoning, and task generalization
• Safety, robustness assessment, and reliable evaluation methodologies for real robots
• Infrastructure for continuous deployment, hardware-software integration, and fleet learning
• Distributed experimentation and runtime diagnostics for real-world robot execution systems

Submission Details

Submissions will be managed through OpenReview and reviewed using a double-blind process. We welcome research papers, system and infrastructure reports, benchmark or evaluation papers, work in progress, and negative or failure-case studies that provide practical insight for the community.

This is a non-proceedings workshop. Accepted submissions will not be included in official proceedings, and we welcome submissions that present or extend previously published work for discussion with the workshop community.

• Submission site: OpenReview submission page.
• Format: IROS / IEEE two-column conference template.
• Length: No more than 8 pages, including references.

Accepted submissions will be presented through selected paper talks or interactive poster presentations. Three submissions have been selected for oral presentation.

Oral Presentations

1. HumanEgo: A Zero-Instrumentation Data-to-Deployment Pipeline for Robot Learning from Human Egocentric Video
2. HoMMI: Learning Whole-Body Mobile Manipulation from Human Demonstrations
3. XPolicyLab: A Unified Standard and Open Ecosystem for Robot Policy Evaluation and Deployment

Poster Presentations

1. VLAFactory: A Unified, Data-Centric Pipeline for Vision-Language-Action Model Development with Per-Stage Evaluation
2. FalconApp: A Mobile Client-Server Pipeline for Auto-Labeled Object Perception
3. A Recipe for Efficient Sim-to-Real Transfer in Manipulation with Online Imitation–Pretrained World Models
4. CRAFT: Video Diffusion for Bimanual Robot Data Generation
5. Robot Skill Compiler: An Auditable Workbench for Language-Specified MuJoCo Environments
6. AffordanceVLA: A Vision-Language-Action Model Empowering Action Generation through Affordance-Aware Understanding
7. AffordanceWAM: Affordance-Aware Joint World–Action Modeling for Robot Manipulation
8. Delay-Aware Diffusion Policy: Bridging the Observation–Execution Gap in Dynamic Tasks
9. Anatomy of a Closed-Loop Collapse: A Causal Case Study of a Compressed VLA Policy
10. Same Demonstrations, Different Outcomes: Staging Failure-Recovery Data in VLA Fine-Tuning
11. EGPO: Entropy-Guided Policy Optimization for VLA Post-Training
12. EAI²: Hardware-Grounded Closed-Loop Evaluation Infrastructure for Embodied AI Development
13. BlenDAgger: Blended Shared Control for Interactive Imitation Learning
14. EpisodeLedger: Content-Addressed Dataset Snapshots for Incremental Robot Learning
15. Dexterous Robot Manipulation from Human Demonstrations via Contact-Anchored Retargeting and Residual Policy Learning
16. ManiTwin: Scaling Data-Generation-Ready Digital Object Dataset to 100K
17. Closed-Loop Physics-Based Manipulation Evaluation at a Commodity Price: What $3.50 Buys You
18. Humanoid Badminton: Learning Dynamic Racket Skills from Limited Human Motion Data
19. Bench2Arena: Semantics-Preserving Agentic Compilation of Embodied Benchmarks into Isaac Lab-Arena
20. Code Agents Automate the Two Manual Bottlenecks of Sim-Based Robot Learning: Digital Twin Scene Construction and Demonstration Collection