MoRAL: Sensor-Grounded BEV Reasoning for Compact VLMs toward Edge-Oriented Autonomous Driving

Ambarish Govindarajulu Kaliamurthi; Kaikai Liu

Publications

Paper and thesis on MoRAL. Presented at IEEE IMC 2026 in Fukuoka. Google Scholar lists the thesis twice, so you may see three rows.

MoRAL: Sensor-Grounded BEV Reasoning for Compact VLMs toward Edge-Oriented Autonomous Driving

Ambarish Govindarajulu Kaliamurthi, Kaikai Liu · IEEE IMC 2026, Fukuoka, Japan (presented) · arXiv:2608.02449

NVIDIA Cosmos Reason 2, trained in two stages: first it learns to read a bird's-eye view we render from LiDAR and radar, then it learns driving reasoning (what is approaching, whether to brake). Presented at IEEE IMC 2026 in Fukuoka.

Bird's-eye-view sensor plot: ego vehicle at center, LiDAR points and tracked objects around it — Bird's-eye view of a driving scene used in MoRAL

arXiv PDF Google Scholar

MoRAL: Multimodal Reasoning for Autonomous Language Models with Sensor-Grounded Spatial BEV Rendering

Ambarish Govindarajulu Kaliamurthi · M.S. thesis · San Jose State University · May 2026 · Advisor: Dr. Kaikai Liu

Thesis that developed the MoRAL pipeline: physics-encoded BEV from LiDAR and radar, two-stage LoRA fine-tuning of Cosmos-Reason2-2B, and edge evaluation on nuScenes.

SJSU ScholarWorks Google Scholar