Skip to content

Some problem about close-loop code #15

@duck-no-bee

Description

@duck-no-bee

"Great work! I am currently reproducing this project. However, I encountered an error while running the closed-loop simulation code. The logs show:

Metric Engine failed with: invalid poses passed to get_route()

I am using a custom script betop.sh (details attached below) to trigger the simulation. Apart from the paths and environment variables, I haven't made any significant modifications to the original code.

Could this be related to a version mismatch in the nuplan-devkit or specific map/dataset dependencies? Any insights would be greatly appreciated."

betop.sh:
cwd=$(pwd)
CKPT_ROOT="$cwd/checkpoints"
export PYTHONPATH="$(pwd):$PYTHONPATH"
export RAY_tmp_dir="/mnt/files3/nuplan_tmp"
export NUPLAN_DATA_ROOT="/mnt/files3/nuplan/dataset"
export NUPLAN_MAPS_ROOT="/mnt/files3/nuplan/dataset/maps"
export NUPLAN_EXP_ROOT="/mnt/files3/nuplan/exp/Betop"

CKPT_PATH="/mnt/files3/nuplan/exp/Betop/exp/training/BeTop/2026.01.15.23.09.43/checkpoints/last.ckpt"

PLANNER="BeTop"
SPLIT='mini'
CHALLENGES="closed_loop_nonreactive_agents closed_loop_reactive_agents"

for challenge in $CHALLENGES; do
echo "Running challenge: $challenge"

python /home/tianyi/code/icml/BeTop-nuplan-fixed/planning/run_simulation.py \
    +simulation=$challenge \
    planner=$PLANNER \
    scenario_builder=nuplan_challenge \
    scenario_filter=$SPLIT \
    experiment_uid="$SPLIT/$PLANNER" \
    distributed_mode='SINGLE_NODE' \
    number_of_gpus_allocated_per_simulation=0.15 \
    enable_simulation_progress_bar=true \
    verbose=true \
    planner.imitation_planner.planner_ckpt="$CKPT_PATH"\
    worker=sequential

done

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions