Sim-to-sim and sim-to-real
What a designer must do
- Freeze the policy interface and diff it across backends [1], [2], [3].
- Confirm the same task trains in both engines [4].
- Treat a successful sim-to-sim transfer as a possible first step toward sim-to-real deployment [5].
- Fit the model to hardware logs. Do not hand-tune it [6].
- Report the gap per joint in position, velocity and torque [7].
- Classify each gap before you pick a fix [8].
- Test on hardware through a written plan [9].
Scope: the sim-to-sim and sim-to-real guidelines. Treat transfer as two stages. Sim-to-sim tests that a policy does not exploit one solver [10], and a success can be a first step toward deployment [5]. Sim-to-real tests the model against your hardware [9].
Gate A: sim-to-sim interface
All rows are from the Isaac Lab 3.0.0 docs, Transfer Policies Between PhysX and Newton. Every field in the table must match exactly, or you are testing a different task.
Gate A: sim-to-sim interface| Level | Must match | Check | Source |
|---|
| must | Actions: term order, joint names, width, target type, scale, offset, clipping | lint: diff resolved action configs | [1] |
| must | Observations: groups, term order, width, history, units, frames, clipping, corruption | lint: diff resolved observation configs | [2] |
| must | Policy state: normalization statistics, recurrent state shape and reset, commands | review: bundle holds the statistics | [11] |
| must | Timing: physics dt, decimation, policy period, action hold. Only Newton substeps may differ within a policy period | lint: diff timing fields | [3] |
| must | Mechanism: body and joint order, active joints, mimic or equality links | lint: compare ordered name lists | [12] |
| must | Episode: reset and command sampling, reward and termination meaning, length, success test | lint: diff the episode config | [13] |
| must | A physics preset must not silently change policy-facing terms. If it does, it is a different task | lint: diff MDP terms across presets | [14] |
| must | Branched robots: one joint and body ordering convention for training and all playback | lint: print joint order per backend | [15] |
Gate B: sim-to-sim test
Gate B: sim-to-sim test| Level | Requirement | Check | Source |
|---|
| should | Step one identical fixed state in each backend over several resets, before training or transfer | sim: record object motion, contact count, gripper effort, penetration, success | [16] |
| should | The same task trains in both engines | sim: train once on each | [4] |
| should | Expect similar behaviour, not identical trajectories | sim: compare task metrics | [17] |
| guide | Treat a successful transfer as a possible first step toward sim-to-real deployment | sim: evaluate the checkpoint in a second backend | [5] |
| should | Reproduce a Newton-only failure at its first failing step: one environment, fixed seed and reset state, no randomization, the same actions in both backends | sim: store this reproduction | [18] |
| should | Log how often actuators saturate and action signs flip | sim: both, per joint | [19] |
| should | Do not port solver settings by number. A value means something only within its solver | review: each backend has its own tuning record | [20] |
Doc for these rows: Isaac Lab 3.0.0 docs, in the Transfer Policies, Prepare an Asset and Solver Differences pages.
Failure signatures
Failure signatures| Symptom | Suspect first | Source |
|---|
| Locomotion policy falls within a few dozen steps in every environment | Joint order, before contacts or friction | [21] |
| Grip force differs between backends at one command | Doubled effort on PhysX, when both mimic fingers have nonzero drives and receive one logical command | [22] |
| Position policy alternates saturated commands | It exploits one solver's response. Raise damping | [10] |
| Friction behaves differently after a port | Backend values do not map one-to-one | [23], [24] |
| Policy shakes violently on hardware | Unmodeled delay | [25] |
| Small commanded motions are not reached on hardware | Stiction not modeled. The policy learned that commands are always reached | [26] |
| Motions land off target after a bench change | A moved fixture invalidates the layout check. The typing policy cannot adjust to a moved keyboard | [27], [28] |
Gate C: system identification
Fit the model to your robot [6]. Give each parameter physically justified bounds [29].
Gate C: system identification| Level | Doc as the guideline states it | Requirement | Check | Source |
|---|
| should | MuJoCo 4.0.0 docs, Modeling | Fit armature, damping, friction, gains, inertia and delay to hardware logs with optimization | measure: command and sensor logs, then mujoco.sysid | [6] |
| must | MuJoCo 4.0.0 docs, Python | mujoco.sysid uses box-constrained least squares, so give each parameter real bounds | review: bounds are physically justified | [29] |
| should | MuJoCo 4.0.0 docs, Python | Use the physically consistent 10-D pseudo-inertia helper for link inertia | sim: assert a valid inertia after fitting | [30] |
| should | MuJoCo 4.0.0 docs, Python | Include sensor and actuator delays and biases when they exist | measure: fit with and without, compare held-out error | [31] |
| guide | mujoco.sysid README, main branch, captured 2026-11-08 | Budget one full rollout per parameter perturbation | sim: time one rollout, multiply by parameter count | [32] |
Use the direct negative solref format in identification models [33]. Combine identification with randomization, not one or the other [34]. Retune damping after increasing armature [35].
Delay
Delay| Level | Requirement | Check | Source |
|---|
| should | Measure control loop timing on hardware. It differs from simulation | measure: timestamp commands, compute jitter | [36] |
| must | MuJoCo delay at or below timestep times nsample. Beyond that it extrapolates non-causally with no runtime error | lint: on each actuator and sensor | [37] |
| should | The smallest effective positive MuJoCo delay is one timestep | sim: step a test signal | [38] |
| should | For stochastic delay or identification, set nsample from the largest delay | lint: nsample is at least the largest delay divided by the timestep, rounded up | [39] |
| should | For continuous-valued quantities in delay identification, use linear or cubic interp to avoid zero-order-hold discontinuity. Not for integer-valued sensors | sim: residual is continuous across a sweep | [40] |
| guide | mjlab replays delay quantized to the physics timestep, so a fine delay needs a fine step | review: resolution against measured delay | [41] |
Friction
Friction| Level | Requirement | Check | Source |
|---|
| should | Plot a real step response to quantify stiction, then add joint friction. The UR10e showed about 0.25 degrees of offset | measure: steady-state offset | [42] |
| should | Model joint friction where stiction exists | sim: train with and without, compare real error | [26] |
| guide | Compare sim and real video of contact while tuning, without waiting for full training | measure: same grasp on both | [43] |
Actuators
Actuators| Level | Requirement | Check | Source |
|---|
| must | Do not assume an ideal actuator model transfers on series elastic drives. On ANYmal, ideal and analytical models both failed | measure: predicted against measured torque | [44] |
| should | Suspect unmodeled delay when a policy shakes violently | measure: command-to-response delay per joint | [25] |
| guide | Consider a learned model when hand-tuned models stall. The ANYmal team spent over a week without much success | measure: torque error per model | [45] |
| guide | Actuator network: joint torque from position error and velocity history | measure: held-out logs | [46] |
| should | Check independence before one per-actuator network. Shared hydraulic supply may couple actuators | measure: neighbours active and idle | [47] |
| should | Model friction, backlash and thermal effects where they matter. Typical motor models lack them | measure: step responses cold and warm | [48] |
Gap metrics
Gap metrics| Metric | How | Source |
|---|
| Paired logs | Replay one motion file on the robot and in sim | [49] |
| Per-joint table | Position, velocity and torque, not one robot-wide number | [7] |
| Error statistics | RMSE, MAPE, correlation, cosine similarity per joint | [50] |
| Steady-state offset | Step response on the real joint | [42] |
| Delay | Command-to-response per joint | [25] |
| Loop timing | Timestamp commands, compute jitter | [36] |
| Dimensions | Key link lengths on the built robot against CAD | [51] |
| Mass | Weigh each removable link | [52] |
| Task success | A stated trial count per task, with the simulation figure beside it | [53] |
| Simulation figure | Quote only with its training-time condition | [54] |
SAGE computes the error statistics, but check it first. Its README badges and requirements name Isaac Lab 2.2.0 and Isaac Sim 5.0.0 [55]. The catalog's release records list Isaac Lab v3.0.0 and Isaac Sim v6.2.0 as current. Confirm its data units too; the state log lists joint data in radians [56]. The full metric list and the measured SO-101 data are in Sim-to-real metrics.
Reference figures
These are one paper's results, on its tasks, not targets [54].
Reference figures| Figure | Value | Condition | Source |
|---|
| Real success, Squint on SO-101 | 91.3 percent | Averaged over 80 trials for 8 tasks | [53] |
| Simulation success, Squint | 96.1 percent | Over all tasks, at the end of its training run | [54] |
Gate D: classify, then fix
Log each gap with one class [8]. The class names come from the guideline. The placement of other guidelines under them is this book's.
Gate D: classify, then fix| Class | Typical check | Source |
|---|
| Sensing | Noise against the simulated model; camera assignment | [57], [58] |
| Actuation | Friction, backlash, thermal, delay | [48], [25] |
| Physics | Contact behaviour against video | [43] |
| Modeling | CAD against as-built, mass against scale | [51], [52] |
Gate E: hardware test plan
Gate E: hardware test plan| Level | Requirement | Check | Source |
|---|
| must | Test each policy through a systematic plan. Good simulation results do not show that it works on hardware | measure: run a written test plan | [9] |
| should | Check input, system response and output consistency separately | review: report covers all three | [59] |
| must | Apply the same policy-output post-processing during real-world inference as in Isaac Lab | review: diff both code paths | [60] |
| should | When a policy fails on the real robot, match the initial observations to simulation and compare responses. This separates observation mismatch from physics or controller mismatch | measure: same observation sequence on both | [61] |
| should | Classify each gap as sensing, actuation, physics or modeling | review: gap log | [8] |
| should | Prefer a simple real controller interface. It has fewer parameters that can mismatch | review: compared against simpler options | [62] |
On the SO-101
On the SO-101| Level | Requirement | Doc as the guideline states it | Source |
|---|
| guide | First transfer: consider scaling actions down and raising control rate. Squint used 0.15 and 30 Hz | Squint, arXiv:2602.21203; SO-101 in ManiSkill3 | [63] |
| should | Record any deliberate change between training and hardware rate | Squint | [64] |
| guide | 10 Hz delta joint position is Squint's training setting, chosen for faster training | Squint | [65] |
| should | Report real success with its trial count | Squint | [53] |
| guide | Cite a simulation figure only with its condition | Squint | [54] |
| must | Calibrate each arm before running a policy | Isaac Lab 3.0.0 docs, SO-101 typing project (patched 3.0.0-based runtime) | [66] |
| must | Typing: keep the keyboard fixed. The policy uses a fixed A–Z key map without camera input and cannot adjust to movement | Same | [27] |
| must | Typing: confirm on hardware that the arm presses and lifts off the right keys | Same | [67] |
| must | Use the typing project's pinned runtime. Unmodified 3.0 GA or later is not validated | Same | [68] |
| should | Moving a fixture or device invalidates the layout check | NVIDIA SO-101 sim-to-real course, latest alias, no version shown, captured 2026-11-08 | [28] |
| must | Verify camera assignments after any replug. Indices can change | Same course | [58] |
| should | Keep layout, lighting and props consistent across calibration, teleoperation and evaluation | Same course | [69] |
Servo values to bound the fit, for the STS3215-C018 at 12 V: the database holds the maker-wiki no-load speed and backlash limit, and bench values for no-load speed, backlash and repeatability. This site does not show them. Take the bounds from the sheet of the servo you buy, or measure them on your arm. The follower uses 1/345 gearing [70].
Where the database is thin
- No guideline gives a general numeric pass threshold for any gap metric. Set your own before the test, and record it. The project-specific gates are listed in chapter 27.
- No part fact gives STS3215 rotor inertia, armature, viscous damping, Coulomb friction or bus delay. Published SO-101 joint fits now exist as guidelines. Their shoulder pan armature differs: 0.1088 [71] against 0.058228342618297214 [72]. The check on the second fit is to run system identification on your arm and compare [72]. Do not mix values from the two fits in one model. Fit your own logs [73].
- The only SO-101 real success rates are from Squint, which used ManiSkill3, not Isaac Lab or MuJoCo. NVIDIA's typing project adds one supervised physical trial, which is not a success rate [74], [75]. MuJoCo Playground reports tests on 3 robot setups, with no success figure in the database [76].
- No guideline describes an Isaac Sim to MuJoCo transfer path. The sim-to-sim guidelines cover PhysX and Newton inside Isaac Lab.
Sources
Each guideline in this chapter with its source, the date it was read and the exact words of the source. Level: must means breaking it causes failure or harm, should is good practice, guide is a rule of thumb.
- Match action term order, joint names, width, target type, scale, offset and clipping exactly between backends. Level: must. isaac-sim.github.io 2026-11-08“Term order, ordered joint or body names, width, target type, scale, offset, and clipping.”
- Match observation groups, term order, width, history, units, frames, clipping and corruption exactly between backends. Level: must. isaac-sim.github.io 2026-11-08“Group and term order, tensor width, history length, units, frames, clipping, and corruption behavior.”
- Match physics dt, decimation, policy period and action hold; only Newton substeps may differ inside one policy period. Level: must. isaac-sim.github.io 2026-11-08“Physics dt, decimation, policy period, and action-hold behavior.”
- Confirm that the same task trains in both engines before attempting transfer. Level: should. isaac-sim.github.io 2026-11-08“ensure that the same task can be trained successfully in both engines.”
- Treat successful sim-to-sim transfer as a possible first step toward sim-to-real deployment. Level: guide. isaac-sim.github.io 2026-11-08“Successful transfer can be a first step toward sim-to-real deployment.”
- Fit armature, damping, friction, gains, inertia and delay to hardware logs with optimization, rather than hand-tuning, when matching a specific robot in MuJoCo. Level: should. mujoco.readthedocs.io 2026-11-08
- Report the gap per joint in position, velocity and torque, not as one robot-wide number. Level: should. docs.nvidia.com 2026-11-08
- Classify each observed gap as sensing, actuation, physics or modeling before choosing a fix. Level: should. docs.nvidia.com 2026-11-08
- Test each policy on hardware through a systematic plan; good simulation results do not show that it works there. Level: must. docs.nvidia.com 2026-11-08
- Raise damping when a position policy alternates saturated commands to exploit one solver's response. Level: should. isaac-sim.github.io 2026-11-08“Increased damping is often necessary to prevent bang-bang control.”
- Carry normalization statistics, recurrent state shape and reset, and commands unchanged with the checkpoint. Level: must. isaac-sim.github.io 2026-11-08“Normalization statistics, recurrent-state shape and reset, commands, and privileged inputs”
- Keep the mechanism the same between backends: body and joint lists in the same order, the same active joints and the same mimic or equality links. Level: must. isaac-sim.github.io 2026-11-08“Ordered bodies and joints, active degrees of freedom, and mimic or equality coupling.”
- Keep the episode definition the same between backends: how resets and commands are sampled, what rewards and terminations mean, episode length and the success test. Level: must. isaac-sim.github.io 2026-11-08“Reset and command distributions, reward and termination meanings, horizon, and success”
- Do not let a physics preset silently change policy-facing terms; if it does, treat the result as a different task. Level: must. isaac-sim.github.io 2026-11-08“A physics preset should not silently change policy-facing action, observation, reward, termination, command, or reset terms.”
- On branched robots, pick a single joint and body ordering convention and use it for the training run and for all playback, on either backend. Level: must. isaac-sim.github.io 2026-11-08“Set both joint_ordering and body_ordering to one convention for training and for every playback”
- Before any training or policy transfer, step one identical fixed state in each backend over several resets and compare the results. Level: should. isaac-sim.github.io 2026-11-08“Run the same fixed task state through both backends before training or transfer:”
- Expect similar behaviour, not identical trajectories, when moving a policy between physics backends. Level: should. isaac-sim.github.io 2026-11-08“Expect similar behavior, not identical trajectories.”
- Reproduce a Newton-only failure at its first failing step with one environment, fixed seed and reset state, no randomization, and the same actions in both backends. Level: should. isaac-sim.github.io 2026-11-08“Reproduce the first failing step with one environment, a fixed seed and reset state, no domain randomization, and the same action sequence in both backends.”
- During transfer, log how often actuators saturate and how often action signs flip between steps. Level: should. isaac-sim.github.io 2026-11-08“monitor saturation and consecutive action sign changes.”
- Do not port solver settings by number, because a configuration value has meaning only within its solver. Level: should. isaac-sim.github.io 2026-11-08“A configuration value has meaning only within the solver that consumes it.”
- If a transferred locomotion policy falls within a few dozen steps in every environment, rule out joint order before you blame contacts. Level: should. isaac-sim.github.io 2026-11-08“Rule ordering out before attributing such a collapse to contacts, friction, or actuator response.”
- On PhysX, check for doubled gripper effort when both mimic fingers have nonzero drives and receive one logical command. Level: must. isaac-sim.github.io 2026-11-08“effectively applying the command twice relative to MJWarp’s single driven finger.”
- Do not assume PhysX friction randomization carries over: Newton events use one friction coefficient, and PhysX values and buckets do not map one-to-one. Level: must. isaac-sim.github.io 2026-11-08“PhysX static/dynamic values and buckets do not map one-to-one.”
- Do not copy PhysX friction numbers into MJWarp; verify shapes and contact first, then fit friction to measured slip. Level: must. isaac-sim.github.io 2026-11-08“Do not copy PhysX friction settings numerically.”
- Suspect unmodeled delay when a policy shakes violently on hardware. Level: should. arxiv.org 2026-11-08
- Model joint friction where stiction exists, or the policy learns that commanded small motions are always reached. Level: should. isaac-sim.github.io 2026-11-08“the policy learns to expect that commanded joint positions are always reached.”
- Keep the SO-101 typing keyboard fixed: its policy uses a fixed A–Z key map without camera input and cannot adjust to movement. Level: must. isaac-sim.github.io 2026-11-08“it cannot adjust to a moved keyboard.”
- Moving any fixture or device invalidates the layout check, so verify positions again before the next run. Level: should. docs.nvidia.com 2026-11-08
- Expect mujoco.sysid to use box-constrained least squares, so give each parameter real bounds. Level: must. mujoco.readthedocs.io 2026-11-08
- Use the physically consistent pseudo-inertia helper when you identify link inertia with mujoco.sysid. 10-D pseudo-inertia Level: should. mujoco.readthedocs.io 2026-11-08
- Include sensor and actuator delays and biases in the identification when they exist. Level: should. mujoco.readthedocs.io 2026-11-08
- Budget one full rollout per parameter perturbation when planning a mujoco.sysid fit. Level: guide. raw.githubusercontent.com 2026-11-08
- To set stiffness and damping independently in MuJoCo, use the direct negative solref format. Level: should. mujoco.readthedocs.io 2026-11-08
- Combine identification with randomization; Hwangbo notes that accurate modeling is cumbersome alone and robustness alone can cost performance. Level: should. arxiv.org 2026-11-08
- Retune damping after increasing armature, because the effective natural frequency and damping ratio change. Level: should. isaac-sim.github.io 2026-11-08“Retune damping after increasing armature because the effective natural frequency and damping ratio change.”
- Measure control loop timing on hardware, because it differs from simulation. Level: should. docs.nvidia.com 2026-11-08
- Keep MuJoCo delay at or below timestep times nsample, because beyond that it extrapolates non-causally with no runtime error. Level: must. mujoco.readthedocs.io 2026-11-08
- Remember that the smallest effective positive MuJoCo delay is one timestep. Level: should. mujoco.readthedocs.io 2026-11-08
- For stochastic delay or identification, set nsample from the largest delay, and adjust actuator_delay at run time within it. Level: should. mujoco.readthedocs.io 2026-11-08
- For continuous-valued quantities in delay identification, use linear or cubic interp to avoid zero-order-hold discontinuity; not for integer-valued sensors. Level: should. mujoco.readthedocs.io 2026-11-08
- Know that mjlab replays actuation delay quantized to the physics timestep, so a fine real delay needs a fine timestep. Level: guide. arxiv.org 2026-11-08
- Plot a real step response to quantify stiction, then add joint friction in simulation; the UR10e showed about 0.25 degrees of offset. 0.25 degrees Level: should. isaac-sim.github.io 2026-11-08“observed contact offsets of approximately 0.25 degrees from the commanded setpoint.”
- Compare sim and real video of contact behaviour while tuning friction and solver values, without waiting for full training. Level: guide. isaac-sim.github.io 2026-11-08“Compare the gear’s behavior in the gripper visually between sim and real”
- Do not assume an ideal actuator model transfers on series elastic drives: on ANYmal, both ideal and analytical models failed. Level: must. arxiv.org 2026-11-08
- Consider a learned actuator model when hand-tuned analytical models stall; the ANYmal team spent over a week without much success. Level: guide. arxiv.org 2026-11-08
- Train an actuator network to predict joint torque from position error and velocity history, when analytical modeling is not good enough. Level: guide. arxiv.org 2026-11-08
- Check the independence assumption before one per-actuator network, because shared hydraulic supply may couple actuators. Level: should. arxiv.org 2026-11-08
- Model friction, backlash and thermal effects where they matter, because typical motor models lack them. Level: should. docs.nvidia.com 2026-11-08
- To measure the actuation gap, replay the same motions on hardware and in simulation and compare the two logs. Level: should. docs.nvidia.com 2026-11-08
- Use RMSE, MAPE, correlation and cosine similarity per joint when comparing with SAGE output. Level: guide. raw.githubusercontent.com 2026-11-08
- Compare CAD against the as-built robot as part of gap measurement. Level: should. docs.nvidia.com 2026-11-08
- This guideline is held back: its source does not allow us to publish it yet.
- Report real success with its trial count, as Squint does with 91.3 percent over 80 trials on 8 tasks, and show the simulation figure beside it. Level: should. arxiv.org 2026-11-08
- Cite a simulation success figure only with its training-time condition: Squint's 96.1 percent is measured at the end of its training run. Level: guide. arxiv.org 2026-11-08
- Check simulator versions before using SAGE, because its README badges and requirements name Isaac Sim 5.0.0, not 6.2.0. Level: must. raw.githubusercontent.com 2026-11-08
- Check units in SAGE data files, because the state log format lists joint data in radians. Level: should. raw.githubusercontent.com 2026-11-08
- Model sensor noise and delay in the observations where they matter for the robot. Level: should. isaac-sim.github.io 2026-11-08“Sensor noise and delays should be modeled appropriately”
- Verify camera assignments after any replug, because indices can change. Level: must. docs.nvidia.com 2026-11-08
- Check input, system response and output consistency separately before you call a transfer sound. Level: should. isaac-sim.github.io 2026-11-08“Successful sim-to-real transfer requires addressing three fundamental aspects:”
- Apply the same policy-output post-processing during real-world inference as in Isaac Lab. Level: must. isaac-sim.github.io 2026-11-08“is also applied during real-world inference”
- When a policy fails on the real robot, match the initial observations to simulation and compare responses to separate observation mismatch from physics or controller mismatch. Level: should. isaac-sim.github.io 2026-11-08“set up the real robot with the same initial observations as in simulation”
- Prefer a simple real controller interface, because it has fewer parameters that can mismatch between sim and real. Level: should. isaac-sim.github.io 2026-11-08“Have fewer parameters that can mismatch between sim and real”
- On a first SO-101 transfer, consider scaling actions down and raising control rate at deployment; Squint used 0.15 and 30 Hz and reports smoother paths. 0.15 action scale Level: guide. arxiv.org 2026-11-08
- Record any deliberate change between training and hardware control rate; Squint accepts such a mismatch and states it. Level: should. arxiv.org 2026-11-08
- Treat 10 Hz delta joint position control as Squint's SO-101 training setting, chosen for faster training. 10 Hz Level: guide. arxiv.org 2026-11-08
- Calibrate each SO-101 before running a policy; another setup's released offsets do not calibrate your robot. Level: must. isaac-sim.github.io 2026-11-08“they do not calibrate another robot.”
- Before deploying SO-101 typing, confirm on hardware that the arm presses and lifts off the right keys. Level: must. isaac-sim.github.io 2026-11-08“checks that the real robot reaches and releases the intended keys in the trained scene.”
- Use the SO-101 typing project's pinned Isaac Lab runtime, because unmodified 3.0 GA or later is not validated. Level: must. isaac-sim.github.io 2026-11-08“Compatibility with unmodified 3.0 GA or later versions has not been validated.”
- Keep the real workspace layout, lighting and props consistent across calibration, teleoperation and evaluation. Level: should. docs.nvidia.com 2026-11-08
- The SO-101 leader uses several gear ratios, while the follower uses 1/345 gearing. Level: guide. huggingface.co 2026-11-08
- The Playground v1.0.0 SO-101 armature values identified from real teleop data are 0.1088 shoulder pan, 0.1472 shoulder lift, 0.1224 elbow, 0.1133 wrist flex, 0.1146 wrist roll and 0.1985 gripper. Level: guide. raw.githubusercontent.com 2026-11-09“# real teleop data. ARMATURE = { "shoulder_pan": 0.1088, "shoulder_lift": 0.1472, "elbow_flex": 0.1224, "wrist_flex": 0.1133, "wrist_roll": 0.1146, "gripper": 0.1985,”
- The gtc_berlin fit gives the SO-101 shoulder pan stiffness 48.93682087435154, damping 2.309705389862818 and armature 0.058228342618297214. Level: guide. raw.githubusercontent.com 2026-11-09“stiffness: 48.93682087435154 damping: 2.309705389862818 armature: 0.058228342618297214”
- Fit 24 parameters for the SO-101: armature, friction loss, stiffness and damping on each of the 6 joints, from teleoperation logs by rollout tracking error. 24 parameters Level: should. raw.githubusercontent.com 2026-11-09“The optimizer fits a 24-parameter model (`armature`, `frictionloss`, `stiffness`, `damping` for all 6 joints) from teleoperation MCAP logs by minimizing rollout tracking error in MuJoCo physics.”
- One supervised trial on a second SO-101 typed NEWTON in 7.10 s; this is one trial, not a success rate. Level: guide. raw.githubusercontent.com 2026-11-09“jointly fitted scene completed **NEWTON** in 7.10 seconds with matching releases and model-FK clearance,”
- The recorded physical result states that one NEWTON trial is not a success rate benchmark or an independent reproduction. Level: guide. raw.githubusercontent.com 2026-11-09“One operator-supervised NEWTON policy trial on the development setup; not a success-rate benchmark or independent-user reproduction.”
- The SO-101 pick-cube policy was tested on 3 robot setups; a visual gap can remain with darker shadows, so light the scene to avoid them. Level: should. raw.githubusercontent.com 2026-11-09“We tested on 3 different robot setups under various lighting conditions, but there may still be visual domain gap with darker shadows.”