Rate of change of rangefinder sensor measurement

[Jan-Loettgen]

Intro

Hi!

I am a researcher at the University of Glasgow, I use MuJoCo for my research on in orbit robotics and reinforcement learning.

My setup

MuJoCo 3.3.0, Python

My question

I am developing a controller, for which I need the rate of change of the distance measured by a rangefinder sensor. I was intending to measure this velocity and acceleration, via numerical differencing. The trouble is that I run many (100-1000) mujoco steps, per controller step, such that the simulation might run at 1000+ Hz while my controller runs at 10 Hz, I do this via:

mujoco.mj_step(m, d, 100)

As such I am really asking whether there is a better way to find the velocity and acceleration than:

mujoco.mj_step(m, d, 98)
sd_pprev = d.sensordata.copy()
mujoco.mj_step(m, d, 1)
sd_prev = d.sensordata.copy()
mujoco.mj_step(m, d, 1)

v_prev = (sd_prev  - sd_pprev)/dt
v_cur = (d.sensordata - sd_prev)/dt

a_cur = (v_cur -v_prev)/dt

because in the test I did I have found this to be considerably slower than the original. I have also considered, taking the sensor data from 100 steps ago, but I suspect this will give singificantly less accurate velocity measurements and also mean that the I get the first acceleration measurement only after 200 steps. Is there any better way to do this that I am not considering? Any advice, or comments appreciated!

Minimal model and/or code that explain my question

No response

Confirmations

• I searched the latest documentation thoroughly before posting.
• I searched previous Issues and Discussions, I am certain this has not been raised before.

[yuvaltassa]

Did you look at mujoco.rollout? It's much faster than Python and will give you all the intermediate sensor values.

https://mujoco.readthedocs.io/en/latest/python.html#rollout

[Jan-Loettgen]

This looks really useful. I am wondering how much speed I will gain over the python approach above, when I only have one simulation. I will do some testing. Thanks