Robotics Engineer · CMU MRSD '26
Deepam Ameria
Robot Learning & Autonomous Systems · Open to Roles
About Me
Researcher. Engineer.
Builder of robots that think.
I'm completing my MRSD at Carnegie Mellon (GPA 3.99), graduating May 2026. Currently a Research Assistant at CMU's Bot Intelligence Group under Dr. Jean Oh, building RIO, a real-time robot I/O framework for teleoperation and VLA policy deployment across robot embodiments (RSS 2026).
Most recently shipped: Lunar ROADSTER, the full autonomy stack for a lunar-grading rover under Dr. Red Whittaker, achieving 89% crater detection across 100+ field hours.
I'm into robot learning, embodied AI, VLA policies, and getting learned behaviors working reliably on real hardware. Open to full-time roles from May 2026.
Featured Projects
Selected Work
Current research, completed capstone, and model-based control. All with real robots.
01 / Current Research
Multimodal Pedestrian Intent Prediction
Text-guided vision transformers for autonomous driving safety
- Building a multimodal intent prediction framework on PIE dataset (300k+ frames, 1,842 pedestrians), fusing visual features with text-based intent prompts
- Designing text-guided attention to ground cues like gaze direction, posture, and head orientation in symbolic traffic context
- Benchmarking against PIEint, I3D, SF-GRU, MaskPCPA baselines; testing zero-shot generalisation on VisionTrap edge cases
02 / MRSD Capstone
Lunar ROADSTER
Full autonomy stack for a lunar-grading rover · Dr. William "Red" Whittaker
- Built the full autonomy stack: depth perception, SkyCam localisation, and multi-sensor state estimation in ROS2 on NVIDIA Jetson
- Trained YOLOv8 crater detector achieving 89% high-confidence detection across 100+ hours of field testing
- Built an RGB-D geometric pipeline to estimate crater geometry and plan autonomous grading trajectories during live field trials
03 / Optimal Control
Model-Based Control for Humanoid Soccer
QP + trajectory optimization for target-directed kicking on Unitree G1
- Developed model-based control for humanoid soccer kicking in Unitree G1 simulation, integrating whole-body planning with ball-strike execution
- Formulated a QP for optimal foot impulse and contact point; used Direct Collocation to generate feasible kicking trajectories
- Implemented IHLQR and TVLQR for balance and trajectory tracking throughout the kicking motion
Experience
Latest Role
From legged robot autonomy to industry and research experience in robot learning, autonomy, and EV systems, across three countries.
- Building RIO, a real-time Robot I/O framework for unified sensor ingestion, teleoperation, and policy execution across robot embodiments (RSS 2026)
- Built high-frequency teleoperation pipelines (50-80 Hz) for large-scale demonstration collection on single-arm and bimanual systems
- Evaluating VLA policies (pi0.5, GR00T) on real robots; extending the framework to Unitree G1 humanoids
- Designed multi-modal teleoperation interfaces (VR, SpaceMouse, master-slave) for scalable human-in-the-loop data acquisition
Skills
Technical Toolkit
Languages
- Python
- C++
- MATLAB
- Bash / Shell
ML & AI
- PyTorch
- Deep Learning / Deep RL
- Behavior Cloning / IL
- Transformer Architectures
Robotics & Middleware
- ROS / ROS2
- OpenCV · Computer Vision
- Sensor Fusion (IMU, GNSS, LiDAR)
- Control Systems
Optimization & Control
- Kalman / EKF Filtering
- LQR, MPC, QP, PID
- Trajectory Optimization
- Direct Collocation
Systems & Hardware
- Docker · Git · Linux
- NVIDIA Jetson
- SolidWorks / Fusion 360
- 3D Printing & Prototyping
Robot Platforms
- Unitree G1, B2, Go2
- Franka Panda (7-DOF)
- Clearpath Husky A300
- Custom mechatronic systems
Core Toolkit
MATLAB
ROS / ROS2
SolidWorks
Education
Academic Background
Carnegie Mellon University
Selected Coursework
- Deep Learning & Deep Reinforcement Learning
- Multimodal Machine Learning
- Computer Vision
- Optimal Control & Robot Autonomy
- Systems Engineering for Robotics
- Space Robotics Development · Medical Robotics
- TA: Mechatronic Design
K.J. Somaiya College of Engineering
Selected Coursework
- Robotics and AI
- Deep Learning in Autonomous Vehicles
- Mechatronics & Electric Vehicles
- Programming in C
Contact
Let's Build Something Together
I'm graduating in May 2026 and actively looking for full-time roles in robot learning, embodied AI, and autonomous systems. Whether you want to discuss research, collaboration, or opportunities, my inbox is open.
Reach me at dameria@andrew.cmu.edu