ANR Robot™

We are a small R&D lab working to advance the state of the art for autonomous robots in unstructured environments. A secondary goal is to explore just how far a small company, research group, or hobbyist can go building on open-source software and off-the-shelf hardware.

Research Overview

🤖 All experiments use the Trossen AI Stationary Robot
pi0.5 policy for puttting a bead on a string, after 2 iterations of policy improvement.
pi0.5 policy for closing a tie wrap, after 1 iteration of policy improvement.
High Dexterity Tasks
ACT policy rollout for popping lid.
ACT policy rollout for pouring little red box.
pi0.5 policy improved by augmenting datasets with human interventions.
Not all results are this good, but still much better than pre-DAgger!
pi0.5 learns a multi-prompt sub-task policy, e.g. 'pick up pink cube ...'
Gemini Robotics ER-1.5 HL controller chooses subtasks to 'put all cubes in bucket'
pi0 LoRA finetuned policy from SIMULATED dataset containing only red cubes.

Hierarchical Planning

Recursive Tree Planner
Builds a stack of 5 blocks in Box2d
Recursive Tree Planner
Solves Lunar Lander in Box2d
(Problem from Gymnasium)
Recursive Tree Planner
Solves Inverted Pendulum in Mujoco
(Problem from Levy et al)
Recursive Tree Planner
Solves 3 Level Four Rooms
(Problem from Levy et al)
Paper: Pure Planning to Pure Policies and In Between with a Recursive Tree Planner