Jaxolotl: A Unified High-Performance Benchmark Suite for LTL-Based Multi-Task RL
Jaxolotl standardizes LTL-based multi-task RL testing and reports speedups of up to 220x.
The paper introduces a JAX benchmark suite with six representative algorithms, four environments, curated task suites, and a statistically robust evaluation protocol. Its static-array handling of symbolic task representations allows fully JIT-compiled training and evaluation. The authors use it to compare existing methods, finding that non-myopic general methods weaken as proposition counts grow, while better-scaling methods depend on environment-specific assumptions and become myopic. Source: ArXiv · AI/CL/LG's note.
The paper introduces a JAX benchmark suite with six representative algorithms, four environments, curated task suites, and a statistically robust evaluation protocol. Its static-array handling of symbolic task representations allows fully JIT-compiled training and evaluation. The authors use it to compare existing methods, finding that non-myopic general methods weaken as proposition counts grow, while better-scaling methods depend on environment-specific assumptions and become myopic. Source: ArXiv · AI/CL/LG's note.
score 5