MemoryBench provides benchmark tasks across spatial memory and action recall categories for robotic manipulation. It serves as the evaluation foundation for the SAM2Act+ framework, focusing on the integration of visual foundation models with memory architectures.
Use Cases
- Evaluate robotic agents on spatial memory tasks using the action recall benchmarks
- Test memory architectures for their ability to maintain spatial awareness across manipulation sequences
- Benchmark visual foundation models like SAM2 in the context of sequential robotic manipulation
- Analyze the performance of robotic manipulation models using the provided task descriptions
Strengths
- Benchmarks spatial memory and action recall in robotic manipulation environments
- Designed for the evaluation of the SAM2Act+ framework and its memory architecture
- Includes task descriptions for testing visual foundation model integration
- Accompanied by open-source code for the SAM2Act framework at github.com/sam2act/sam2act