Business Problem
Offline metrics did not reflect safety-critical behavior well enough, making it harder to compare autonomy models before real-world testing.
+13% correlation improvement
IROS 2025
First-author research
AI Solution
Animikh designed an evaluation method incorporating prediction uncertainty and used foundation vision models to improve Sim2Real transfer from simulation to real-world environments.
Outcome
+13% improvement in ground-truth correlation and first-author research accepted at IROS 2025.
Technical Shape
The implementation focused on production constraints rather than demo-only wins: architecture, data realities, evaluation, inference behavior, deployment path, monitoring, and maintainability.
PyTorchCARLASegment AnythingDepth AnythingAutonomous driving
He independently comes up with strong solutions to challenging research problems while keeping up with recent advancements in AI and computer vision.
Client or collaborator testimonial excerpt
Related Services
Where this work maps into AVIC Labs offers
Build a system with this level of care.
Send a short project brief and AVIC Labs will respond with the right next step.
Email AVIC Labs