Robotics Researcher, Evaluations

About the company

Efference provides high-performance perception and compute infrastructure for training and deploying robot models in the cloud. Our vision is to make robotic deployments scalable with shared compute.

We approach the problem by owning the entire path from the image sensor to the datacenter GPU. We shrink the load at the sensor, fix the uplink so robot traffic is fast and reliable, and clock synchronize the system so observations arrive time-aligned, ready for inference.

About the role

You'll determine how changes in Efference's perception and compute infrastructure affect what robots can actually do. You'll design experiments that connect camera quality, compression, timing, network behavior, and inference performance to reliable task completion on physical robots.

What you’ll do

• Build a suite of manipulation and mobile-robot tasks with clear success criteria, repeatable starting conditions, and representative failure cases.
• Integrate and evaluate learned policies, including vision-language-action and imitation-learning models, across local and remote inference configurations.
• Run controlled studies of bitrate, image processing, observation age, synchronization error, latency, and jitter, separating their effects from changes in the policy.
• Measure task success, intervention rate, completion time, and recovery behavior with enough repeated trials to distinguish improvements from noise.
• Maintain versioned datasets, checkpoints, environments, and run metadata; use held-out tasks or scenes to test whether results generalize.
• Investigate failures with imaging, wireless, controls, and ML engineers, then turn the findings into product requirements and reproducible regression cases.

What we are looking for

• Hands-on experience deploying and evaluating learned policies on physical robots.
• Strong Python skills and familiarity with a modern ML framework, robot middleware, and the practical debugging of perception-to-action systems.
• Understanding of experimental design, statistical uncertainty, dataset leakage, and the limits of aggregate benchmark scores.
• Ability to isolate failures across sensor input, model behavior, control, and hardware instead of attributing every failure to the model.
• A record of turning research questions into experiments that others can reproduce and use to make engineering decisions.

Nice to Have

• Experience with VLA, diffusion, or flow-based policies and robot learning datasets or evaluation frameworks.
• Work on robustness, distribution shift, sim-to-real transfer, or policy performance under sensor and network impairments.
• Experience building automated robot test stations, ground-truth instrumentation, or large-scale evaluation pipelines.