Static DAG Compilation for Runtime-Efficient AV Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicle prediction systems face inefficiencies in determining object properties as they must reconfigure complex directed acyclic graphs (DAGs) at runtime, consuming computational resources and energy, which can impact prediction processes.
Innovation Solution
Building DAGs at compile-time, where nodes are ordered by depth to ensure dependencies are updated before the node itself, allowing for efficient execution of property determination during runtime without reconfiguration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If DAGs are configured at runtime to determine object properties, then the system can adapt to different detection scenarios, but computational resources and energy are consumed during runtime reconfiguration
Solution Approach 1:
The patent pre-compiles the directed acyclic graphs (DAGs) at system initialization or offline stages, organizing the property determination logic and dependencies before runtime. This preliminary action stores the optimized execution sequences in memory, eliminating the need for computational reconfiguration during runtime object property determination, thus reducing energy consumption while maintaining adaptability through pre-loaded scenario-specific DAGs
2Measurement precision
If complex DAGs are reconfigured at runtime, then accurate property determination can be achieved, but computational resources are consumed and prediction efficiency is impacted
Solution Approach 1:
The system performs preliminary compilation of complex DAGs into optimized execution sequences stored in memory. During runtime, the pre-compiled sequences are directly executed without reconfiguration, maintaining accurate property determination through the preserved dependency relationships while significantly improving prediction efficiency by eliminating runtime compilation overhead
Solution Approach 2:
The patent merges the DAG compilation and execution processes by storing the compiled representation in memory and executing it directly. This merging eliminates the separation between compilation and runtime execution, allowing the system to achieve both accurate property determination (through preserved DAG structure) and high prediction efficiency (through eliminated reconfiguration steps)
3Ease of manufacture
If nodes are processed in arbitrary order, then implementation is simpler, but dependency requirements may not be met and accuracy is compromised
Solution Approach 1:
The system performs preliminary topological sorting during DAG compilation to establish the correct execution order of nodes based on their dependencies. This pre-computed ordering is stored with the DAG structure, allowing the runtime system to execute nodes in the correct sequence without complex runtime dependency analysis, thus maintaining implementation simplicity while ensuring accurate property determination through proper dependency satisfaction
Data Source
AI summary
At compile-time, a processor develops a computer program by receiving an input that includes multiple nodes and connections between pairs of the nodes. The nodes represent object properties such as properties of objects that an autonomous vehicle (AV) detects while moving about an environment. For each node, the system will identify a depth that represents a number of nodes along a longest path from that node to any available input node. The system will order the nodes by depth in a sequence, and it will build a graph-based program specification that includes the nodes in the sequence, along with the connections. The graph-based program specification may correspond to a directed acyclic graph (DAG). The system will compile the graph-based program specification into a computer-readable program, and it will save the computer-readable program to a memory so that the AV or other system can use it at run-time.