ML Engineer; perception & state estimation

Department

Perception

Job Type

Full-time

Location

Cambridge

Modality

Hybrid

Start Date

No access

We are looking for a Machine Learning Engineer to join our Perception Team . You will build the core perception and reasoning engine for our flagship multi-agent system. This role is responsible for architecting the software that transforms raw, noisy sensor data into a rich, symbolic world model. This team will develop and implement the algorithms for managing perception inputs and maintaining a knowledge manager based on such inputs.

You'll form part of the Perception Team . This team unlocks the mastermind’s understanding and reasoning about its environment.

This is an on-site position, the successful candidate will be expected to work from the office at least 3 days a week.

Multi-Sensor Fusion:
Design and implement algorithms that manage the fusion of heterogeneous sensor streams (e.g., EO/IR, LiDAR, and neuromorphic cameras) into a single, coherent picture of the world.

Object Recognition:
Build and deploy models for real-time object detection, classification, and tracking, transforming raw data into structured, classified objects with unique IDs and states.

World Modeling:
Develop the Knowledge Manager , the central repository for abstract and symbolic world knowledge. You will be responsible for inferring the logical relationships between objects and agents.

Probabilistic State Estimation:
Implement and maintain the belief state over the environment, a core component of a knowledge manager.

Goal Inference:
Create the logic that translates high-level user commands into the formal, predicate-based goal states.

API

Collaboration:

Work closely with the Systems and Behaviour teams to define and refine APIs.

A strong theoretical foundation and practical experience in probabilistic machine learning (e.g., Bayesian inference, Gaussian Processes, state estimation filters like EKFs/UKFs).

Demonstrable experience with modern ML frameworks (PyTorch preferred) and computer vision libraries (OpenCV) applied to real-world sensor data.

Hands-on experience with sensor fusion techniques for combining data from sources like cameras and LiDAR.

Production-quality coding skills in both Python and C++.

Proven experience developing and deploying software for real-world robotic systems (e.g., UAVs, UGVs)

Deep expertise in sensor fusion techniques, particularly with state estimation filters like EKF, for tracking and localization.

Hands-on experience with the Robot Operating System (ROS
2) and an understanding of the underlying DDS middleware and its QoS settings.

Practical experience in multi-agent reinforcement learning (MARL), planning under uncertainty, or collaborative robotics.

Familiarity with high-fidelity simulation environments for robotics, especially NVIDIA Isaac Lab.

Familiarity with the challenges of real-time systems, including managing latency, ensuring deterministic timing (e.g., PTP), and maintaining performance on degraded communication links.

Experience with knowledge representation, logical inference, or symbolic reasoning systems.