Axis Robotics Releases Task 74 Manipulation Dataset with Raw vs. Smoothed Trajectory Comparison
Axis Robotics released Task 74 sample dataset on Hugging Face—a robotics manipulation dataset featuring a Franka arm performing a long-horizon pick-and-pour task. The dataset includes raw web-collected trajectories and processed smoothed versions for downstream training. This appears to be a data quality/intentory release showcasing their data collection pipeline structure, specifically demonstrating the difference between unsmoothed raw trajectories and the smoothed training data used for policy learning. The task encompasses multiple manipulation primitives: grasping, transport, pose control, and pouring.
Axis Robotics Releases Task 74 Manipulation Dataset with Raw vs. Smoothed Trajectory Comparison
Axis Robotics published a sample dataset on Hugging Face demonstrating a Franka Panda arm executing a kettle-to-mug pouring task. The dataset's key contribution is its side-by-side comparison of raw web-collected trajectories and their smoothed counterparts used for downstream policy training—a practical reference for understanding data preprocessing pipelines in robot learning.
Integration Strategy
When to Use This?
This dataset serves educational and reference purposes rather than direct application:
- Research Reference: Understanding data preprocessing requirements for manipulation tasks
- Pipeline Development: Evaluating trajectory smoothing approaches for custom datasets
- Benchmark Comparison: Assessing baseline trajectory quality for similar manipulation tasks
- Curriculum Development: Training material for robot learning courses
Relevant Domains:
- Domestic service robotics
- Kitchen automation
- Long-horizon manipulation research
- Imitation learning data engineering
How to Integrate?
Access Method:
- Hugging Face Hub repository (link provided in source)
- Sample dataset (not full-scale production dataset)
- Download format: Standard Hugging Face dataset structure
Integration Steps:
# Conceptual integration (actual API may vary)
from datasets import load_dataset
dataset = load_dataset("axis-robotics/task74-sample")
raw_trajectories = dataset["raw"]
smoothed_trajectories = dataset["smoothed"]
No public API documentation confirmed—researchers should inspect the repository structure directly.
Compatibility
Expected Specifications (Inferred):
- Format: Likely NumPy arrays or PyTorch tensors within Hugging Face dataset format
- Framework: Compatible with PyTorch, JAX, and any framework supporting HF datasets
- Environment: Assumed OpenAI Gym or MuJoCo-style simulation (standard for Franka Panda)
No explicit compatibility information confirmed for:
- Specific simulation environments (Isaac Gym, RaiSim, etc.)
- ROS versions
- Real-world deployment requirements
Source: Hugging Face via Axis Robotics Reference: Axis Robotics Task 74 Sample Dataset on Hugging Face Hub DevRadar Analysis Date: 2026-04-29