Sorry for the larger request, these are features developed along risk-constrained paper:
Safe distance evaluators ( static, dynamic, for agents and corridor)
Improve Frenet calculation by integrating orientation of shapes with respect to center line. The lateral distance is now between shapes and not between centerline and coordinate of other vehicles. This allows for a more meaningful parameter to select when vehicles are considered as front vehicle.
BenchmarkRunner:
fixes
merge from file: Large files caused out of memory when merging checkpoints. Merging at file level uses copy operations to solve this issue
EvaluationConfig allows for scenario type specific evaluation configs
Python-Based Evaluators specified in other modules
RandomMacroActions behavior is meaningful for testing scenarios or as a baseline
Scaling and Enlarging Functions for polygons with separate scaling along x and y
Serialization of road corridor lane ids to support intersection scenarios in the database
Drawing:
of lane polygons in viewer
evaluations results during benchmarking
camera settings are parameterizable (supporting scenario dependent camera settings directly in the json file)
Sorry for the larger request, these are features developed along risk-constrained paper: