SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Pickle Robot is automating global supply chains with Physical AI, deploying robots to unload trucks and containers in warehouses. The company is building toward a generalized autonomy engine (Dill) to orchestrate robots across entire logistics processes.
You will own machine learning problems end-to-end for the perception systems that control how robots grasp packages. This is a senior individual contributor role focused on the modeling loop: problem formulation, data strategy, training and experimentation, and rigorous evaluation. Platform and robotics engineers handle runtime integration and edge deployment, so you stay focused on ML.
Key responsibilities:
- Own ML problem formulation: translate picking performance goals into well-posed ML problems, defining what the model predicts, how labels and rewards are defined, loss functions, and success metrics.
- Build the data engine: own dataset strategy end-to-end, including large-scale synthetic scene generation, distributed cloud dataset jobs, scene diversity filtering, and self-supervised label generation. Identify and fix data bottlenecks.
- Train and iterate on large vision models: fine-tune large backbones with multiple task-specific heads, run disciplined experiments, and know when to stop.
- Own evaluation and verification: define offline metrics that predict on-robot outcomes and online metrics for edge case improvement.
- Drive technical excellence: conduct rigorous code reviews, mentor junior engineers, and contribute to the perception and grasping roadmap.
- Hand off cleanly to production: produce deployable models with sane latency budgets, exportable graphs, and reproducible artifacts.
Requirements:
- 5+ years of experience in machine learning and computer vision, with models you personally carried from problem statement to measured outcome.
- Track record of framing problems: translate high-level product requirements into specific engineering tasks and explain technical trade-offs to non-experts.
- Hands-on experience building datasets at scale: generation or collection, curation, automated quality assurance, coverage and bias analysis, active learning.
- Rigorous evaluation habits: baselines, ablations, metric design, clear sense of when a result is real.
- Expert-level Python and PyTorch skills.
- Comfortable with cloud training infrastructure (GCP or AWS), Docker, and experiment tracking tools.
- Professional experience with at least one of: multi-task models, multi-modal models, multi-view models, or self-supervised/outcome-based/reward-driven learning.
