Disentangling perception and reasoning for improving data efficiency in learning cloth manipulation without demonstrations

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• The paper's modular approach to RL for cloth manipulation significantly reduces model size and training time compared to end-to-end learning methods.method 85%
• The proposed method achieves significant performance improvements over available baselines on the SoftGym benchmark for cloth manipulation tasks.result 90%
• End-to-end learning approaches for cloth manipulation incur significant computational cost by training agents on highly lossy representations of environment state.limitation 80%
• The paper demonstrates successful sim-to-real transfer of simulation-trained cloth manipulation models to the real world.technical 85%
• Reinforcement learning is considered a promising approach to cloth manipulation problems because analytical methods have not provided robust and general manipulation policies.technical 90%
• Data-based methods for cloth manipulation usually rely on large models and long training times, resulting in high computational costs.limitation 85%
• The paper questions the common design choice of end-to-end learning with workspace images as input for cloth manipulation policies.method 90%
• Cloth manipulation presents challenges due to high-dimensional state space, complex dynamics, and high propensity to self-occlusion in fabrics.technical 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card