Unverified50% confidenceBenchmarkExact time
In the 'object hanging' task of the MAGIC project, traditional model-based planning took over 5 minutes for a single solve, while a representative method based on unsupervised feature learning and policy learning, after training on 100,000 synthetic cup data, achieved only a 42% single-demonstration generalization success rate; this framework, by using visual correspondences to guide constrained optimization, raised the success rate to 93% from just a single demonstration.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
UnverifiedAdapt-1's reactive policy mode starts in an untrained state, accumulates experience through approximately 36 minutes of actual gameplay to form a policy, then disables learning and directly uses the acquired policy to complete the level.68% similarUnverifiedDuring fine-tuning, model weights, gradients, and optimizer states must be stored simultaneously, making VRAM consumption typically 3–4 times that of inference.68% similarUnverified小模型通过知识蒸馏通常比直接在标注数据上训练精度高出2-5个百分点68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1024130API
curl https://kongchang.com/api/v1/knowledge/claims/1024130MCP
get_claim(id=1024130)