Cannot build Open3D with CUDA in Docker

Hello everybody !

First of all, a big thanks for the people who will spend some times for answering to this post.
I am currently trying to create a docker container with CUDA enable for a specific application with an IFM 3D camera (which run on Ubuntu 18.04 with CUDA 10.2 and Nvidia runtime).
The application is running in CPU mode, but to enable the GPU (CUDA) we need to build the Open3D library from source.

To do that, I have tested two containers, one from “l4t-cuda:10.2.460-runtime” and the second one from “l4t-base:r32.4.3”.
Both finish with the same error code during CMake process :
Unable to find cudart library.
CMake Error with the function FindCUDAToolkit => Could NOT find CUDAToolkit (missing: CUDA_CUDART) (found version “10.2.300”).
I really don’t know what can I do and why I have this error message with the base container from Nvidia directly :(

My Dockerfile :


USER root

# Installation de GIT
RUN apt-get update && \
    apt-get install -y git sudo && \
    apt-get clean

# Installation de CMAKE
RUN apt update && \
    apt install -y software-properties-common && \
    apt clean all && \
    wget -O - 2>/dev/null | gpg --dearmor - | tee /etc/apt/trusted.gpg.d/kitware.gpg >/dev/null && \
    apt-add-repository "deb bionic main" && \
    apt update && \
    apt install -y cmake

# Téléchargement et installation de Open3D
RUN git clone --depth 1 --branch ${OPEN3D_VERSION} ${OPEN3D_CLONE_REPOSITORY} && \
    cd Open3D && \
    util/ assume-yes && \
    mkdir build && \
    cd build && \
    cmake -DCUDAToolkit_ROOT=/usr/local/cuda-10.2/lib -DCMAKE_INSTALL_PREFIX=/usr \
        -DPYTHON_EXECUTABLE=/usr/bin/python3 \
        -DBUILD_GUI=OFF \
        -DOPEN3D_ML_ROOT= \
        .. && \
    make -j$(nproc) && \
    make install && \
    make -j install-pip-package