observie Profile Banner
David Bar Profile
David Bar

@observie

Followers
2K
Following
4K
Media
213
Statuses
1K

A software guy building robots.

Joined February 2024
Don't wanna be here? Send us removal request.
@observie
David Bar
7 hours
Goodbye gen1 hardware, welcome gen2 😘
Tweet media one
2
1
44
@observie
David Bar
7 hours
(gen2 still in prototype stage but already better).
0
0
4
@observie
David Bar
2 days
@devemin you may find this useful.
1
0
1
@observie
David Bar
8 days
#6: PA12+GF (Glass Fiber reinforcements). Not sure if this is a finalist, probably not, but this material at least passed the first round. No major failures :). BUT! Keep this mind when considering:
Tweet media one
1
0
7
@observie
David Bar
8 days
#5: Aluminum Leg. It's heavy, expensive and takes time to manufacture, but it won this round. Looks beautiful and the performance is great.
Tweet media one
1
0
10
@observie
David Bar
8 days
#4: Slim PA12 structure. A slim limb structure that looks and works great in aluminum was surprisingly warped and wobbly in PA12.
Tweet media one
1
0
3
@observie
David Bar
8 days
#3: HP-PA12 threads do not work. I've manufactured four robotic legs in four different materials: Aluminum, PA12, HP-PA12 and PA12-GF. In terms of threading, HP-PA12 was the worst. Consider using thread inserts or just skip this material? It does look nice though 🙂
2
0
6
@observie
David Bar
8 days
#2: early PLA prototypes. Early PLA prototyping is fast and fun, but getting it anywhere close to a heating motor is a sure way to create Dali-inspired robotic art.
Tweet media one
Tweet media two
3
1
11
@observie
David Bar
8 days
In software development, we live by Donald Knuth's "premature optimization is the root of all evil" approach: start simple, profile, then optimize. I believe it applies to hardware too - but boy - when software breaks, it just means rewriting a few lines, whereas hardware
Tweet media one
5
5
112
@observie
David Bar
9 days
🤣
Tweet media one
0
0
1
@observie
David Bar
9 days
RoboDog: first push-up (the robot's Hello World?)
9
7
117
@observie
David Bar
9 days
@grok manufacturing inconsistencies is not a great recipe for a smooth sim2real transfer.
0
0
4
@observie
David Bar
9 days
Another GIM6010-8 gets quite hot while idle: 51.4°C (124.3°F). Others are reasonable, some even quite cool, but overall thermal performance among the 12 motors is inconsistent (see attached 12 measurements and graph by @grok). At the very least, redundantly hot motors mean the
Tweet media one
Tweet media two
@observie
David Bar
20 days
One GIM6010-8 motor gets crazy hot just sitting idly: around 50°C (122°F) for doing nothing. All others remain calmly cool.
Tweet media one
5
0
7
@observie
David Bar
10 days
The internals of a Unitree Go2 leg. Beautiful engineering.
@toodooda
Takayuki Todo / 藤堂高行
24 days
まずこのGo2を修理しなければならないのだ。.(先週末についに骨折した).なお上海・民生現代美術館での展覧会が3週間後。
Tweet media one
2
0
18
@observie
David Bar
10 days
An interesting approach to PD tuning for RL from @ChongZitaZhang:. - very high P gain, in the hundreds.- let RL do the magic.- 50Hz control loop. Result:. Thank you Chong Zhang ❤️.
@ChongZitaZhang
C Zhang
10 days
@observie very big gains, hundreds. RL does the magic.
0
0
8
@observie
David Bar
11 days
My AI friend wants me to read these two, claiming that "Åström/Murray's "Feedback Systems" kicks off with killer basics on PID + FF for actuators like yours, showing how velocity/torque FF cuts errors in joint positioning without PID strain—super relevant for smooth quadruped
Tweet media one
Tweet media two
3
1
24
@observie
David Bar
11 days
Pure gold from @KyleMorgenstein . Setting soft PID gains would allow an RL policy to achieve compliance, if needed, and when not, the policy can always exaggerate commands (e.g. by setting position values beyond their targets), and then rapidly adjust, based on realtime feedback,.
@KyleMorgenstein
Kyle🤖🚀🦭
11 days
@observie for RL you want to use gains so soft it couldn’t support the robot’s weight under gravity. for model based control double the P gain. you should not have to adjust the PD params at runtime. run the PD controller as fast as you can (I’ve seen up to 40kHz) and interpolate commands.
2
3
24
@observie
David Bar
11 days
Impressions after spending a Saturday with RoboDog. 1. Motors' behavior under load is a whole new ball game. 2. A position command that easily sends an unloaded motor to the desired position behaves completely differently when motors are loaded. 3. This is amplified when.
@observie
David Bar
13 days
RoboDog is standing for the first time!. Look daddy, no pillows 🙂. All 12 motors are online, periodically sending their position estimates via CAN. A control user space Rust application is monitoring the motor positions, and sending them position commands.
Tweet media one
5
1
48