IPAB Workshop - Thursday 15th October @ 2pm

Speaker: Servinar Cheema

Title: Investigating the effectiveness and behavioural affordances in scenario-based black-box testing of automated driving systems.

Abstract: Simulation-based testing validates black-box agents by generating scenarios to test for failures under a given specification. However, current pipelines often decouple generation from specifications, flatten metrics, and rely on lagging indicators alone. We applied a mutation-based testing approach and found that modern simulation-based testing toolchains did not fully detect the injected driving faults. We propose a preprocessing step that couples scenario generation with specifications in an active test loop. By constructing a solution space of valid behaviours from kinematic and behavioural constraints before scene creation, we obtain known and preserved compliance bounds. This shifts evaluation from retrospective scoring to leading metrics based on proximity to failure boundaries.

Speaker: Jakub Zadrozny

Title: Bootstrapping Articulated 3D Reconstruction from 2D Image Collections

Abstract: 3D reconstruction of articulated objects from a single image is challenging because large training datasets with paired image and 3D supervision are difficult to obtain. Recent point map-based methods achieve strong performance but rely on synthetic datasets rendered from manually created articulated 3D assets with carefully curated pose distributions. While camera viewpoints can be easily sampled, generating realistic object articulations remains costly and labor-intensive. We propose a training framework that reduces this requirement by leveraging unannotated 2D images collections with only a single rigged canonical mesh per category. Starting from a weak 3D shape predictor trained on canonical-pose renders, we iteratively estimate object articulation and camera pose by fitting the mesh to predicted point maps. The recovered articulations and viewpoints are then used to render updated synthetic training data, progressively improving the predictor. Despite using substantially weaker 3D supervision, our models achieve performance comparable with DualPM, which requires manually curated articulated training datasets.