Robotics & Embodied AI — Inference Phase

🤖

Inference Specialization

Robotics & Embodied AI

Embodied AI engineers build systems where robots perceive the world and act in it. Massive 2025 investment: Figure AI, Boston Dynamics, Tesla Optimus, 1X — over $6.8B raised. Vision-language-action models are transforming the field rapidly.

Days

Projects

$195K

Median

$6.8B

2025 Investment

$140K–$260K

Salary Range

Week 1: ROS 2, Simulation, and Robot Kinematics · Days 1–10  

Set up the standard robotics software stack, simulate environments in 3D, and understand robot motion and coordinate systems.

D1–3

ROS 2: nodes, topics, services, and TF coordinate transforms

ROS 2 architecture, DDS middleware, TF transforms, Nav2 navigation stack, and launch systems

↗ ROS 2 Humble official tutorials (free)

D4–6

Simulation: NVIDIA Isaac Sim, Gazebo, and PyBullet

Sim-to-real transfer methodology, domain randomisation, and photorealistic training environments

↗ NVIDIA Isaac Sim tutorials (free)

D7–9

Robot kinematics, dynamics, and trajectory planning

Forward and inverse kinematics, Jacobians, trajectory planning with MoveIt2, motion constraints

↗ Modern Robotics Specialization — Northwestern/Coursera (free audit)

🏗 Project — Day D10: Simulated Navigation Robot 

· ROS 2 navigation stack with Nav2 stack configured

· LiDAR-based obstacle avoidance behaviour

· Occupancy grid map building from simulated sensor data

· Path planning demonstration video in simulation

Week 2: Perception, Reinforcement Learning, and Imitation · Days 11–28  

3D perception from real sensors, RL for motor control, and learning precise manipulation from human demonstrations.

D11–14

3D perception: point clouds, depth cameras, and SLAM

LiDAR processing, PointNet++, RGB-D cameras, simultaneous localisation and mapping algorithms

↗ Self-Driving Cars Specialization — UoT/Coursera (free audit)

D15–18

Reinforcement learning for robot control: PPO, SAC

Reward shaping, sparse rewards, sim-to-real transfer, RL for locomotion and manipulation tasks

↗ Deep Reinforcement Learning — Sergey Levine, Berkeley (free lectures)

D19–22

Imitation learning and behaviour cloning for manipulation

Learning precise manipulation from demonstrations. Diffusion Policy, ACT transformer, data collection workflows

↗ Imitation Learning for Robotics papers — arXiv (free)

D23–27

SLAM, localisation, and multi-sensor fusion

EKF-SLAM, IMU and GPS and LiDAR fusion, robust localisation in dynamic outdoor environments

↗ SLAM Course — Cyrill Stachniss (YouTube, free)

🏗 Project — Day D28: Robotic Arm Manipulation Policy 

· 25 or more demonstration trajectories recorded in simulation

· Behaviour cloning policy achieving over 80% success rate

· Domain randomisation experiments across 3 texture sets

· Sim-to-real transfer gap analysis and recommendations

Week 3: Foundation Models for Robotics and Safety · Days 29–40  

Vision-language-action models transforming general-purpose robotics, safe deployment near humans, and fleet management.

D29–32

VLA models: RT-2, OpenVLA, and pi-zero

Foundation models connecting vision, language, and robot actions. The path to general-purpose robotic intelligence

↗ RT-2 paper and project site — Google DeepMind (free)

D33–36

Safe reinforcement learning and human-robot interaction

Deploying robots near humans. Collision avoidance, safe RL with constraints, certified safe policies

↗ Safety in Robotics — MIT OpenCourseWare (free)

D37–39

Robot fleet management and over-the-air updates

Managing multiple robots at production scale, fleet monitoring, coordinated model updates, failure detection

↗ ROS 2 fleet and multi-robot tutorials (free)

🏗 Project — Day D40: Autonomous Robot System — Full-Stack Capstone 

· Perception pipeline from real camera or simulated LiDAR

· VLA-based high-level task planner

· RL or PD low-level motor controller

· Failure detection and recovery behaviour

· Simulation demo video with narrated walkthrough