- Framework for learning a hand intent recognition model from sEMG for FES-based control. Technical University of Munich, 2024, mehr…
- Online detection of compensatory strategies in human movement with supervised classification: a pilot study. Frontiers in Neurorobotics 17, 2023 mehr…
- Assessing Human-Human Kinematics for the Implementation of Robot-Assisted Physical Therapy in Humanoids: A pilot study. The International Conference on Rehabilitation Robotics (ICORR) 2023, 2023 mehr…
- Deep Learning based Uncertainty Decomposition for Real-time Control. 2023, The 22nd World Congress of the International Federation of Automatic Control, 2023 mehr…
- Safe Learning-Based Control of Elastic Joint Robots via Control Barrier Functions. 2023, The 22nd World Congress of the International Federation of Automatic Control, 2023 mehr…
M.Sc. Neha Das
Technical University of Munich
Chair of Information-oriented Control (Prof. Hirche)
- Phone: +49 (89) 289 - 25729
- Room: 0305.04.520
- Homepage
- neha.das@tum.de
Short Biography
- 10/2020 - present: PhD candidate at Chair of Information-Oriented Control (ITR), Technical University of Munich (TUM), Germany.
- Projects: Part of ReHyb, COMAN, conPDMode
- 2019 - 2020: AI Resident, Facebook, U.S.A.
- Projects: Inverse Dynamics Learning, Inverse Reinforcement Learning
- 2016 - 2019: Masters in Informatics, Technical University of Munich, Germany.
- Projects: Intrinsically Motivitated Control, 3D Semantic segmentation of Human Bodies, Computer Vision, Machine Learning
- 2009 - 2013: B.Sc. in Software Engineering, Delhi Technological University, India.
- Projects: Algorithms, Datastructures, Artificial Bee Colony Analysis (Final Project)
Research Interests
My research interests include
- learning representations of healthy human motor behavior
- data-driven anomaly detection and correction of motor behavior
- learning-based control from user preferences
- learning cost functions via inverse optimal control
Working Field
- H2020 project "Rehabilitation based on Hybrid Exoskeleton" [ReHyb]
Motion health analysis for stroke patients
Stroke survivors often compensate for the loss of motor function in their distal joints by altered use of more proximal joints and body segments. This can be detrimental to the rehabilitation process in the long-term as it hinders regaining normal function of impaired joints in the training. It is important to detect such motor patterns and correct them as soon as possible, either via visual or verbal feedback or through assistive control mechanisms.
Challenges
Detecting these unusual motor patterns can be challenging due to the following issues:
- Context dependency
These behaviors vary significantly across patients and across tasks. A motor behavior labeled compensatory in one context may be healthy in another context! - Sparse data
Only limited amounts of motor behavior data can be obtained from patients! - Label Uncertainty
Experts may not concur on labeling motor behaviors as compensatory or non-compensatory which can lead to high output noise!
Our Goal: Detection and Correction of unhealthy motion patterns
For this we require sensors, that can observe patient motions, actuators, that can influence patient motions and a control algorithm, that can derive the correction actuation given the observations from the sensors. An example of a system of sensor and actuators is the ReHyb Hybrid Exoskeleton that has encoders that can observe patient joint kinematics and can influence patient motions via Torque and FES control.
The aim is then to develop a safe and stable corrective control design for the detection and correction of unhealthy motor behavior.
For Prospective Students
If you are interested in doing a Forshungpraxis and/or a Bachelor's or Master's Thesis under my supervision, then please contact me via Email (neha.das@tum.de), along with your CV and grades.
Currently available Thesis:
| Title | MA | BA | FP | IP |
|---|---|---|---|---|
| Learning Latent Spaces for Probabilistic Movement Primitives [PDF] | X | X |