[{"data":1,"prerenderedAt":100},["ShallowReactive",2],{"\u002Fen\u002Fglossary\u002Fsim-to-real":3},{"id":4,"title":5,"alternateName":6,"body":7,"description":90,"extension":91,"keywords":92,"meta":93,"navigation":94,"path":95,"seo":96,"stem":97,"updated":98,"__hash__":99},"glossary\u002Fglossary\u002Fen\u002Fsim-to-real.md","Sim-to-Real","仿真到实机迁移",{"type":8,"value":9,"toc":84},"minimark",[10,15,22,27,34,62,66],[11,12,14],"h1",{"id":13},"what-is-sim-to-real","What Is Sim-to-Real?",[16,17,18,21],"p",{},[19,20,5],"strong",{}," is the approach of training robot control policies in a physics simulator (such as MuJoCo or Isaac Gym\u002FLab) and then deploying them on real hardware. Training on real robots is slow, expensive, and breaks hardware; simulation runs thousands of robots in parallel at faster-than-real-time speed, which is what makes reinforcement learning of humanoid walking practical at all.",[23,24,26],"h2",{"id":25},"the-core-challenge-the-reality-gap","The Core Challenge: The Reality Gap",[16,28,29,30,33],{},"Simulation never matches reality exactly — friction, latency, motor characteristics, and sensor noise all differ. This mismatch is the ",[19,31,32],{},"sim-to-real gap",". The main mitigations:",[35,36,37,44,50],"ul",{},[38,39,40,43],"li",{},[19,41,42],{},"Domain randomization",": randomly perturb simulation parameters (mass, friction, latency) during training, forcing the policy to become robust;",[38,45,46,49],{},[19,47,48],{},"System identification",": measure the real robot's motor response and inertial parameters accurately and write them back into the simulator;",[38,51,52,55,56,61],{},[19,53,54],{},"Actuator modeling",": model the torque-speed behavior of the ",[57,58,60],"a",{"href":59},"\u002Fen\u002Fglossary\u002Fjoint-motor","joint motors"," explicitly — often the deciding factor for legged-robot transfer.",[23,63,65],{"id":64},"hardware-requirements","Hardware Requirements",[16,67,68,69,73,74,78,79,83],{},"Deployed policies command joints at hundreds of Hz, demanding high-bandwidth force control (see ",[57,70,72],{"href":71},"\u002Fen\u002Fglossary\u002Fquasi-direct-drive","quasi-direct drive"," and the ",[57,75,77],{"href":76},"\u002Fen\u002Fglossary\u002Fmit-protocol-can","MIT protocol",") and low-latency buses. The BXI ",[57,80,82],{"href":81},"\u002Fen\u002Frobots\u002Fhumanoid-robot","Elf 3 humanoid"," ships with a MuJoCo environment and ROS2 SDK, and its >1000 Hz PCIE-CANFD control architecture supports the full simulation-to-hardware workflow.",{"title":85,"searchDepth":86,"depth":86,"links":87},"",2,[88,89],{"id":25,"depth":86,"text":26},{"id":64,"depth":86,"text":65},"Sim-to-Real is the approach of training robot control policies at scale in physics simulation and then transferring them to real robots — the dominant training paradigm for humanoid locomotion today.","md","sim-to-real, simulation to reality, reinforcement learning robotics, domain randomization, MuJoCo, Isaac",{},true,"\u002Fglossary\u002Fen\u002Fsim-to-real",{"title":5,"description":90},"glossary\u002Fen\u002Fsim-to-real",null,"r1zVLzbVErZiDCC98F5liP8bIuQEPdAdUZGonwnDjak",1783425867452]