Autonomy Intern, Perception

LionsBot International Pte Ltd.; Singapore; On-site; SGD 13,000 - 27,000; Full time

Job summary

LionsBot International in Singapore is offering an internship to contribute to our data flywheel pipeline powering our robotics fleet's AI capabilities. You will prototype a dataset retrieval system and run 2D detection/segmentation experiments, producing benchmarking results and actionable recommendations.

The role emphasizes building reproducible research code, benchmarking modern vision models on real-world robot data, and delivering concrete outputs to support production stack decisions.

Responsibilities

• Support the building of the data flywheel pipeline that powers the AI capabilities of our robotics fleet.
• Build a smart database query system for fast dataset retrieval and collation.
• Designing and running visual model experiments for 2D detection and segmentation pipeline, with results provided back to the team to act on.

Job description

Overview

The intern will support development of the data flywheel pipeline powering our robotics fleet's AI capabilities, including a prototype query system for dataset retrieval and collation. They will also design and run 2D detection and segmentation experiments, delivering benchmarking results and recommendations to the team.

Key Responsibilities

• KR1: Support the building of the data flywheel pipeline that powers the AI capabilities of our robotics fleet.
• KR2: Build a smart database query system for fast dataset retrieval and collation.
• KR3: Designing and running visual model experiments for 2D detection and segmentation pipeline, with results provided back to the team to act on.

Deliverables

• From KR2: A working prototype of a smart database query system.
• From KR3: Provide sets of benchmarking results for vision model experiments, with clear findings and recommendations for the production stack.

Learning Objectives

State what the intern should be able to do by the end of the placement. These should be specific and assessable — they will also form the basis of the mid-point and final review.

See the full data flywheel in motion: collection, curation, labelling, QA, and retraining at fleet scale.

Train and benchmark modern vision models on real-world data from robots in live commercial environments.

Put vision-language models to work and automate what used to be manual annotation effort.

Build the engineering habits that separate research code from shipped code (reproducibility, versioning, clean experiments).