Dataset · lzLiang
Robot memory benchmark for long interactive tasks
lzLiang/EmbodiedMemoryBench
2,554 episodes in AI2-THOR and ProcTHOR simulators for evaluating the memory of embodied agents.
What's inside
The benchmark includes 2,554 interactive episodes (totaling 1,000 to 10,000 records) covering four types of memory tests in AI2-THOR and ProcTHOR environments. Data is presented in a multimodal English format: images, tables, and text. The agent first observes the interaction history without knowing the future goal, and then must use the accumulated memory to perform actions.
What you can do with it
- Evaluating long-term memory of multimodal robotic agents
- Testing action planning systems in a changing 3D environment
- Comparing memory architectures for Embodied AI
- Tasks
- Robotics
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 47.2K
- Likes
- 5
How to load it
from datasets import load_dataset
ds = load_dataset("lzLiang/EmbodiedMemoryBench")