Riddhi Chatterjee, Ritish Shrirao, Kushagra Mehrotra, Viswanath Gopalakrishnan
This paper challenges the assumption that better image-editing quality yields better goals in goal-conditioned manipulation policies. The authors propose SGE-Goal, a spatially-grounded decomposed editing framework, which outperforms large proprietary editors in downstream policy success. SGE-Goal achieves state-of-the-art results on CALVIN and SIMPLER-Bridge benchmarks.
The paper's findings have significant implications for robotic manipulation tasks, where the quality of goal images directly affects the success of downstream policies. The proposed SGE-Goal framework can improve the efficiency and effectiveness of robotic manipulation systems. This research contributes to the development of more reliable and autonomous robots.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.