Agent Navigation in Partially Known Directed Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for controlling the movement of a navigating agent in a network with partially represented directed graphs are inefficient as they either prioritize path planning in known areas or excessively explore unknown parts, failing to find optimal paths due to lack of complete network knowledge and not effectively managing exploration costs.
Innovation Solution
A method that iteratively plans a path using a cost heuristic and performs exploration in unknown parts, incorporating a risk assessment to balance navigation and exploration, allowing the agent to dynamically adjust its route and prioritize known terrain while exploring unknown zones, using a modified A* algorithm with additional risk evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If the agent prioritizes path planning in known areas using algorithms like A* or Dijkstra, then route efficiency is improved, but the agent cannot discover optimal paths through unknown areas
Solution Approach 1:
The patent implements dynamic adjustment of exploration vs. exploitation by modifying the heuristic function based on exploration budget and unknown area characteristics. The algorithm dynamically switches between planning mode (using A*/Dijkstra) and exploration mode (using randomized approaches) based on real-time assessment of unknown areas, allowing optimal path discovery while maintaining route efficiency.
Solution Approach 2:
The patent changes key parameters including the heuristic function (adding exploration bonus terms), exploration budget allocation, and transition thresholds between planning and exploration modes. These parameter modifications enable the agent to balance between utilizing known network information and investing resources in discovering unknown areas that may contain shortcuts.
2Adaptability or versatility
If the agent explores unknown parts of the network, then potential shortcuts are discovered, but exploration costs increase
Solution Approach 1:
The patent implements partial exploration by allocating a bounded exploration budget rather than exhaustive exploration of all unknown areas. The agent performs selective exploration of promising unknown regions based on heuristic assessment, avoiding wasteful exploration of areas unlikely to contain valuable shortcuts, thus reducing overall exploration costs while maintaining adaptability.
Solution Approach 2:
The patent incorporates feedback mechanisms where the agent continuously monitors exploration outcomes, updates its assessment of unknown areas, and adjusts future exploration decisions based on past experiences. This feedback loop allows the agent to learn from previous exploration costs and successes, optimizing the balance between exploration investment and route optimization gains.
3Speed
If the agent uses traditional path planning algorithms in partially known networks, then navigation in known areas is efficient, but the agent lacks exploratory capability
Solution Approach 1:
The patent creates a multi-functional navigation system that integrates both efficient path planning (A*/Dijkstra) and exploratory capabilities within a single unified framework. The system can operate in planning mode for known areas, exploration mode for unknown areas, or hybrid mode combining both, making it universally applicable to partially known networks while maintaining both navigation efficiency and exploration capability.
Solution Approach 2:
The patent performs preliminary assessment of unknown areas using heuristic functions before committing to full exploration. By pre-evaluating potential exploration targets and prioritizing those most likely to yield valuable shortcuts, the agent prepares exploration plans in advance, reducing unnecessary exploration actions and maintaining navigation speed while preserving exploration capability.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The invention relates to a method for controlling the movement of a moving agent (16) within an environment (10) partially represented by a directed graph comprising vertices located at the boundary of an unknown portion (U) of the environment, said method comprising guiding the agent (16) by repeating the following series of steps at least once: planning a path (B) to a destination vertex (14) or boundary vertex (F) by gradually increasing evaluation of a first cost heuristic; controlling a movement of the agent (16) towards the destination vertex (14) or boundary vertex (F) sent back by planning; controlling an exploration in the unknown portion (U) if planning sends back a boundary vertex (F) as a destination. The planning step comprises, for each calculation of the first cost heuristic on a boundary vertex (F), an additional calculation of a second risk heuristic. It yields as a destination the boundary vertex (F) on which the evaluation of the calculated risk heuristic is weakest as soon as the cost exceeds this heuristic risk evaluation before the destination vertex (14) is reached.