TECHNICAL ARTICLES
Implementation decisions, experiments and unresolved problems in robotics. Explore our simulation research by topic.
Project scope and results →
· Simulation research
A prototype for greeting, bowing and guiding visitors. From human-body estimation to model-based whole-body control without RL, with development footage, diagrams and remaining limitations.
Read article →
· Simulation research
A KUKA arm and Sharpa hand execute a public RL insertion policy. Results from 3/3 initial environments and 10/10 with another seed, plus normal-speed footage and evaluation limits.
Read article →
· Simulation research
A 1 kg box in a custom Isaac Sim factory. The original condition passed; four lower-friction conditions failed to lift. One trial per condition, with success and failure videos.
Read article →
· Simulation research
Lifting, carrying and placing a box with a public HAIC policy. Three fixed-start trials and ten small placement variations include a three-second hands-off stability check.
Read article →
· Simulation research
Connect fixed-camera tracking to a G1 camera using Visual SLAM, preserve the same person ID, and verify ROS 2 stop reasons and observation history through log replay.
Read article →
· Simulation research
Two 30-second runs at a 3.3 m/s command produce continuous pose output but up to 11.25 m endpoint error. Stereo and IMU results, normal-speed videos and unresolved issues.
Read article →
· Simulation research
Four opposing loads from 0 to 0.6 Nm in Isaac Sim. Rotation and holding met the test criteria, while slip and body orientation remain unresolved. No physical valve operation has been verified.
Read article →
· Simulation research
Three trials per condition compare walking and running. Continuous pose output can still drift by meters. We include fresh-initialization tests and the condition that fell.
Read article →
· Simulation research
Integrating Caltech’s Chasing Autonomy policy into Isaac Sim, investigating initial phase and friction, then measuring speed and foot contact during sustained running.
Read article →
· Simulation research
Additional observations repair a map with unknown areas. G1 follows a Nav2 route around shelves and stops 8.5 cm from the goal in a recorded simulation trial.
Read article →
· Simulation research
A 70-second walk between shelves builds a colored point cloud. We explain the route and visualization changes, seven-target results and a separate saved-map relocalization test.
Read article →
· Simulation research
Command smoothing did not remove leg jitter. Comparing evaluation settings and matched model/policy pairs reduced high-frequency joint-angle variation by 45.6% in the measured test.
Read article →
· Simulation research
When a robot stands but does not move forward, should only the reward change? A review of published B2-W records separates posture, falling, wheel/leg behavior and policy inputs.
Read article →LET’S BUILD TOGETHER
Tell us about the software you need to implement or test. We can discuss the work, the development setup and how we could contribute.
Discuss your development needs