This position is aimed at enrolled students of the TU Dresden.
BASELABS develops data-fusion and object-tracking algorithms in the automotive domain. Currently, the company designs a new framework (BASELABS Create Embedded) for data-fusion systems that are ready for deployment in safety-critical applications. The framework is implemented in C and a specialized configuration language called Trait-C. Trait-C adds advanced features such as type-safe templates and compile-time polymorphism to C.
Grid mappings, or grid-based estimation algorithms are well-known algorithms which have been used successfully for decades in robotics. One application of this algorithm is the estimation of an occupancy grid . The main limitation of this approach is its applicability for static environments only. With a considerable number of dynamic objects in the environment the quality of the estimated world model significantly deteriorates. Recently, Multi-Instance Bernoulli PHD filters (aka Dynamic Grids) and similar techniques have been proposed as alternative approaches for environmental model estimation suitable for highly dynamical scenarios. These methods represent the world around an object as a set of moving particles, which are employed both for the objects’ state and density-occupancy estimation. The estimation results for static and dynamic environments are used as an input for higher-level functionality such as object classification using machine learning algorithms or advanced driver-assistance systems.
The major objective of this task is to conduct a feasibility study on a possible extension of BASELABS Create Embedded for GPGPU hardware (e.g., by Nvidia) with a prototypical implementation of a grid-based estimation algorithm along with its optimization for execution on GPGPU hardware and/or embedded devices with parallel capabilities. The basic algorithm, which is intended to be implemented, is described in pseudo-code in . The study therefore investigates the general possibility of extending BASELABS Create Embedded and the Trait-C language to support algorithms like those presented in .
The individual tasks are as follows:
1. Prototypically implement the algorithm in BASELABS Create Embedded using C and Trait-C. This includes:
- adapting the framework and/or the language to effectively support (1) if necessary,
- implementing several basic blocks from the paper  for parallel architectures using CUDA and OpenCL. The company can provide several reference implementations for key steps which may be needed to be adapted or changed.
2. Benchmark the parallel implementations against each other and a sequential implementation.
3. Investigate possible implications for applications that use MISRA-C constraints as typical in the automotive domain.
- Student in a Computer Science (or related) Master’s or Diploma program at TU Dresden
- Good Bachelor or “Vordiplom” in Computer Science (or related subject)
- Knowledge in C, C++ or C#
- Desirable is experience with compilers or programming GPU’s
The working student job requires personal attendance in our Chemnitz office for 2-3 days / week.
The thesis will be supervised academically at the chair for compiler construction at TU Dresden.