Pranav Kolekar

Problem

Robot validation needed to stay close to production behavior without drifting into a separate simulation-only path.

Solution

Built a reusable runtime integration layer that connects cloud simulation, command execution, and telemetry at a safe abstraction boundary.

Decision

Kept the integration boundary narrow so the system stayed reusable, reviewable, and safe to evolve.

Impact

Improved sim-to-real confidence and reduced duplicate simulation-only logic.

Problem

Multi-repo simulation environments required slow manual VM setup, while cloud scaling behavior was hard to predict.

Solution

Architected a reproducible Brev and Docker Compose workflow for simulation setup, startup, and validation on cloud GPU instances.

Decision

Standardized the environment before optimizing performance, then used benchmark data to separate CPU limits from GPU assumptions.

Impact

Eliminated onboarding friction and gave the team practical evidence for cloud simulation sizing.

Problem

Navigation, docking, manipulation, perception, and placement needed to be validated as one robot workflow.

Solution

Built an end-to-end digital twin workflow for AMR movement, arm coordination, perception-assisted picking, and task execution.

Decision

Validated the full task path instead of presenting isolated demos that would miss integration failures.

Impact

Created a stronger pre-hardware validation loop for mobile manipulation behavior.