Unverified50% confidenceFactExact time
PointMaze is a sparse-reward continuous-control navigation task in which the agent must reach the goal point in a maze to receive a reward, and the observation does not directly give the goal direction.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn embodied navigation, embedding object grounding into the policy as an explicit intermediate representation is equivalent to injecting a structural inductive bias of 'locate the target first, then decide'.68% similarUnverifiedThe scene graph provides dense rewards for each task stage, ensuring that when the policy starts from a reset state, the exploration path is sufficiently short and clearly guided.67% similarUnverifiedTraditional benchmarks have shortcomings for tool-calling agents: they are typically static question-answer pairs and cannot simulate real-world side effects and state changes.66% similarVerifiedThe phenomenon where minor deviations at each step of a multi-step task executed by an agent may amplify along the chain and lead to results far from the initial intent is called goal drift.65% similarUnverified面对目标漂移应强化任务锚定，面对状态漂移应频繁重新观测真实环境，面对策略漂移应引入循环检测和策略切换机制65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088293API
curl https://kongchang.com/api/v1/knowledge/claims/1088293MCP
get_claim(id=1088293)