Hohoo AI Journey

Virtual Embodied Lab

Start embodied AI experiments without a physical robot. Build a verifiable software path, then explore simulation, data and policies.

Try it · Grid planning

Expand to explore in your browser. Link to this exercise ↗

Try it · Robot instructions

Expand to explore in your browser. Link to this exercise ↗

VL01

MuJoCo pick & place

Learning
Simulation engine
MuJoCo
Robot
Teaching XYZ Cartesian gripper; no complete robot arm, IK or ROS2 integration
Environment
Existing teaching scene: a tabletop, cube and target box with a Cartesian gripper; not a full robot arm or ROS2 integration.
Main project
hohoo-embodied-agent →
Linked milestones

Question to validate

Can the arm move the red cube into the blue box while recording control steps and failure causes?

Experiment steps (planned)

  1. Find cube
  2. Move end effector
  3. Grasp
  4. Move
  5. Place
Technology & skills
  • Joint
  • Link
  • DOF
  • Pose
  • Coordinate Frame
  • Joint Space
  • Cartesian Space
  • FK
  • IK
  • Trajectory
  • Gripper
  • Controller

Result

E14 connects completion contracts to recovery: eight rollouts separate historical placement, current validity and execution qualification. Full VL01 and M4/M5 remain open.

Related resources

VL02

ROS2 & Gazebo robot system

Planned
Simulation engine
Gazebo + ROS2
Robot
Virtual robot; base or arm to be selected
Environment
Virtual camera, IMU, LiDAR and joint state; configuration to be validated
Main project
hohoo-embodied-agent →
Linked milestones

Question to validate

Can virtual sensors publish observations through ROS2 and connect perception, planning and control nodes?

Experiment steps (planned)

  1. Virtual sensor
  2. ROS2
  3. Perception node
  4. Planner node
  5. Controller
  6. Virtual robot
Technology & skills
  • Node
  • Topic
  • Publisher
  • Subscriber
  • Service
  • Action
  • Message
  • Parameter
  • Launch
  • rosbag2
  • TF2
  • URDF

VL03

ManiSkill robot learning

Planned
Simulation engine
ManiSkill
Robot
Virtual manipulation arm; model to be selected
Environment
Candidate tasks: PickCube and pick & place; not run yet
Main project
hohoo-embodied-agent →
Linked milestones

Question to validate

Can a manipulation policy learn from robot data and be evaluated for actions and failures in virtual tasks?

Experiment steps (planned)

  1. Observation
  2. Policy
  3. Action
  4. Environment
Technology & skills
  • PyTorch
  • Behavior Cloning
  • Imitation Learning
  • ACT
  • Diffusion Policy
  • RL
  • Policy Evaluation
  • VLA Evaluation

VL04

Isaac Physical AI lab

Future
Simulation engine
Isaac Sim / Isaac Lab
Robot
Arms, mobile robots or humanoids; candidate directions only
Environment
Isaac Sim / Isaac Lab; not installed, hardware and compute to be assessed
Main project
hohoo-embodied-agent →
Linked milestones

Question to validate

Later, study synthetic data, parallel simulation and domain randomization to evaluate policy transfer limits.

Experiment steps (planned)

  1. Synthetic data
  2. Parallel simulation
  3. Domain randomization
  4. Transfer evaluation
Technology & skills
  • Isaac Sim
  • Isaac Lab
  • Synthetic Data
  • GPU Parallel Simulation
  • Domain Randomization
  • Sim2Real

Physical hardware belongs to later validation; it is not a prerequisite for starting.

Later, towards the real world →