Master Thesis: Learned Filtering and Association for Multi-Target Tracking
Ericsson ABNy
Sammanfattning
Join a team at Ericsson focused on enhancing the Bayesian pipeline for tracking multiple targets using advanced filtering techniques. This role involves implementing KalmanNET and a transformer-based DataAssociator within the StoneSoup framework, evaluating performance against various metrics, and fine-tuning components with proprietary data. The position is based in Stockholm, Sweden, and offers a collaborative environment for innovative problem-solving.Det här erbjuder vi
Opportunity to work on innovative solutions to complex problems.Collaborative environment with a diverse team of innovators.Encouragement of diverse and inclusive workplace culture.
Ansök senast: Öppet tillsvidare
Publicerad: 2026-09-22
Beskrivning
Join our Team
About this opportunity:
StoneSoup's Bayesian pipeline (Predictor -> Hypothesiser -> DataAssociator -> Updater) tracks multiple targets using Kalman-family filters (KF, EKF, UKF, CKF) and probabilistic association (GNN, JPDA, EHM). Fixed process and measurement noise Q/R can cause filter divergence under model mismatch, while hand-crafted distance metrics can cause track coalescence and cubic association cost. This thesis replaces two pipeline stages with learned components: KalmanNET as the Updater, adapting Kalman gain K and noise covariances Q/R online via a GRU; and a transformer as the DataAssociator, producing association probabilities via attention instead of fixed gating.
What you will do:
• Implement KalmanNET as a drop-in Updater, learning K, Q, and R from the innovation sequence
• Implement a transformer-based DataAssociator using cross-attention between track and detection tokens
• Build a StoneSoup simulation curriculum: linear to non-linear motion, low to high clutter, manoeuvring and crossing targets
• Train both components on the curriculum with NEES/NLL/MSE loss for the filter and Hungarian-matched CE/GIoU loss for the associator
• Evaluate against KF/EKF/UKF/CKF/IMM and GNN/JPDA/EHM2/TrackFormer using OSPA, MOTA, IDF1, NEES, and latency
• Fine-tune and validate both components on Ericsson proprietary RAN measurement data
The skills you bring:
• Working knowledge of Kalman filtering and Bayesian state estimation
• Python proficiency, including PyTorch
• Familiarity with recurrent networks (GRU/LSTM) and attention/transformers
• Comfort with StoneSoup or similar tracking frameworks
• Basic linear algebra and probability, including covariance, Cholesky decomposition, and Gaussian densities
• Understanding of multi-object tracking metrics such as OSPA, MOTA, and IDF1
• Experience with simulation-based training curricula
• Git-based, reproducible experiment workflow
Why join Ericsson?At Ericsson, you'll have an outstanding opportunity. The chance to use your skills and imagination to push the boundaries of what's possible. To build solutions never seen before to some of the world's toughest problems. You'll be challenged, but you won't be alone. You'll be joining a team of diverse innovators, all driven to go beyond the status quo to craft what comes next.
What happens once you apply? Click Here to find all you need to know about what our typical hiring process looks like.Encouraging a diverse and inclusive organization is core to our values at Ericsson, that's why we champion it in everything we do. We truly believe that by collaborating with people with different experiences we drive innovation, which is essential for our future growth. We encourage people from all backgrounds to apply and realize their full potential as part of our Ericsson team. Ericsson is proud to be an Equal Opportunity Employer. learn more.
Primary country and city: Sweden (SE) || Stockholm
Req ID: 791101
About this opportunity:
StoneSoup's Bayesian pipeline (Predictor -> Hypothesiser -> DataAssociator -> Updater) tracks multiple targets using Kalman-family filters (KF, EKF, UKF, CKF) and probabilistic association (GNN, JPDA, EHM). Fixed process and measurement noise Q/R can cause filter divergence under model mismatch, while hand-crafted distance metrics can cause track coalescence and cubic association cost. This thesis replaces two pipeline stages with learned components: KalmanNET as the Updater, adapting Kalman gain K and noise covariances Q/R online via a GRU; and a transformer as the DataAssociator, producing association probabilities via attention instead of fixed gating.
What you will do:
• Implement KalmanNET as a drop-in Updater, learning K, Q, and R from the innovation sequence
• Implement a transformer-based DataAssociator using cross-attention between track and detection tokens
• Build a StoneSoup simulation curriculum: linear to non-linear motion, low to high clutter, manoeuvring and crossing targets
• Train both components on the curriculum with NEES/NLL/MSE loss for the filter and Hungarian-matched CE/GIoU loss for the associator
• Evaluate against KF/EKF/UKF/CKF/IMM and GNN/JPDA/EHM2/TrackFormer using OSPA, MOTA, IDF1, NEES, and latency
• Fine-tune and validate both components on Ericsson proprietary RAN measurement data
The skills you bring:
• Working knowledge of Kalman filtering and Bayesian state estimation
• Python proficiency, including PyTorch
• Familiarity with recurrent networks (GRU/LSTM) and attention/transformers
• Comfort with StoneSoup or similar tracking frameworks
• Basic linear algebra and probability, including covariance, Cholesky decomposition, and Gaussian densities
• Understanding of multi-object tracking metrics such as OSPA, MOTA, and IDF1
• Experience with simulation-based training curricula
• Git-based, reproducible experiment workflow
Why join Ericsson?At Ericsson, you'll have an outstanding opportunity. The chance to use your skills and imagination to push the boundaries of what's possible. To build solutions never seen before to some of the world's toughest problems. You'll be challenged, but you won't be alone. You'll be joining a team of diverse innovators, all driven to go beyond the status quo to craft what comes next.
What happens once you apply? Click Here to find all you need to know about what our typical hiring process looks like.Encouraging a diverse and inclusive organization is core to our values at Ericsson, that's why we champion it in everything we do. We truly believe that by collaborating with people with different experiences we drive innovation, which is essential for our future growth. We encourage people from all backgrounds to apply and realize their full potential as part of our Ericsson team. Ericsson is proud to be an Equal Opportunity Employer. learn more.
Primary country and city: Sweden (SE) || Stockholm
Req ID: 791101
Ansök till tjänsten
Master Thesis: Learned Filtering and Association for Multi-Target Tracking
Ny
OM FÖRETAGET
Ericsson AB
