FAQ and Troubleshooting

FAQ and Troubleshooting#

Known problems with installing, running and benchmarking Open Apero, each with its cause and fix.

Installation#

rosdep install cannot locate a ROS package#

Cannot locate rosdep definition for [<ROS package>]: rosdep was updated without the end-of-life Noetic distro. Run rosdep update --include-eol-distros, then rosdep install again.

rosdep install cannot locate eigen_catkin, catkin_simple or voxblox#

The repository was cloned without --recursive, so the voxblox fork and its vendored dependencies, which have no rosdep keys, are missing: git -C ~/catkin_ws/src/open_apero submodule update --init --recursive.

CMake stops: OR-Tools is not installed#

open_apero_core links open_apero_core/ortools/lib/libortools.so directly. Copy OR-Tools as in OR-Tools Setup, then run catkin build again.

CMake stops: empy not found under conda#

Unable to find either executable 'empy' or Python module 'em': the package was configured with the python3 of conda’s auto-activated base environment. Run conda deactivate, remove ~/catkin_ws/build/<pkg>, which keeps that python3, and run catkin build again; conda config --set auto_activate false (conda before 25.5: auto_activate_base) keeps base inactive in new shells, but not in the current one.

open_apero_voxblox does not compile, or a voxblox connection is dropped#

A missing getUpdatedMapRegion, or an md5sum mismatch on voxblox_msgs/Layer, means an upstream voxblox is in the workspace or an underlay. Remove it; the fork ships as the deps/voxblox submodule.

roslaunch aborts: HSSD_DATA_ROOT is not set#

The scene files read the dataset location from it. Run source datasets.env (HSSD Dataset).

The Habitat node cannot import habitat_sim#

ModuleNotFoundError: No module named 'habitat_sim': the first line of ~/catkin_ws/devel/lib/open_apero_habitat/habitat_renderer_node names /usr/bin/python3, not the Habitat environment’s python3. Rebuild open_apero_habitat as in Habitat Setup.

The Habitat node exits: unable to find EGL device for CUDA device 0#

habitat-sim’s CUDA build cannot initialize CUDA, so it finds no GPU to render with. Check CUDA with nvidia-smi and, where PyTorch is installed, python3 -c "import torch; print(torch.cuda.is_available())"; if either fails, reload the nvidia_uvm module (sudo rmmod nvidia_uvm && sudo modprobe nvidia_uvm) or reboot.

pip install habitat-sim finds no package#

habitat-sim has no Linux build on PyPI; install 0.2.3 from conda as in Habitat Setup.

Running#

Symptom

Cause and fix

A red REQUIRED process [...] has died! with process has finished cleanly at the end

The normal end of a graph-benchmark sweep or a simulated run: the node is required, so roslaunch exits with it

The planner waits and logs [settle] warnings

The settle gate has not passed; the warning names what is missing. For a sensor whose cloud is not in the camera’s optical frame, such as a LiDAR, set wait_for_map_updated: false (Integrating a Robot)

toggle_running refuses to start the planner

With log_performance on, open_apero_node needs a writable ~run_dir; set it first, as in Start and Stop

<type> ignores parameters: a, b

The plugin’s setup did not read these keys of its namespace: usually a misspelt key or a key of another type; strict_params: true makes it fatal (How Parameters Are Read)

rosparam load leaves $(env …) in a parameter

The configs are roslaunch-templated; load them through the launch files (Configuration Files)

A graph-benchmark dataset does not reproduce after a partial rebuild

A benchmark_node older than libopen_apero_core.so can disagree through a stale vtable: catkin build both packages and rerun (If a Dataset Does Not Reproduce)

A Docker run joins another ROS master#

The container shares the host’s network, so it joins a master already on port 11311. Give the run its own port with roslaunch -p 11411 ... and export ROS_MASTER_URI=http://localhost:11411 for the other ROS tools in that shell. The batch tools pick their own: paper_sweep.py from --master-port-base, run_experiments.sh from SWEEP_MASTER_URI.