watonomous.github.io
[ Knowledge Base : Meetings with Kevin ]
Created by [ Charles Zhang], last modified on Oct 02, 2020
October 2nd
+———+————————————–+——————–+ | Topic | \ | \ | +=========+======================================+====================+ | RADAR | How far is something going to move | \ | | | then update. Both measured and | | | | perceived velocity. Need to convert | | | | measured velocity to perceived | | | | velocity | | | | | | | | If motion characteristics are | | | | seriously different, we may be | | | | forced to use two trackers. Kevin | | | | thinks that one tracker is good | | | | enough. | | | | | | | | Radar Clustering - get from | | | | Kevin. | | | | | | | | Filtering: | | | | | | | | - Stationary, in path, low | | | | signal → send it | | | | - Moving, heading into the path | | | | - Stationary, out of path → filter | | | | it out | | | | - In motion → send it | | | | | | | | Stale: nothing on the track list | | | | | | | | Multiple sensors: Fused | | | | | | | | Mathworks object tracking - look at | | | | this | | | | | | | | Get point cloud from radar | | | | | | | | - do clustering | | | | - Compare to doppler effect to | | | | look for velocity | | | | - Cars near you are going to look | | | | like an L shape | | | | | | | | Pick a Goal: | | | | | | | | - Detect moving Deer, walking | | | | pedestrians | | | | - For now, focus on moving stuff | | | | | | | | Need memory: | | | | | | | | - Lateral motion is the best. | | | | Mount 45 degrees if possible. | | | | | | | | Variable thresholds: if in path, use | | | | lower threshold | | | | | | | | use a series of different filters. | | | | Pedestrian at walking speed might be | | | | different | | | | | | | | Convert from relative speed to | | | | absolute speed | | | | | | | | Radars negotiate so they do not | | | | cause interference | | +———+————————————–+——————–+ | Safety | - Shower curtain | {.content-wrap | | for | - implications of only having two | per} | | in-car | people in the entire environment | Charles | | testing | - 3 people to do a testing | Zhang{.confluen | | | back corner of plot they store univ. | ce-userlink | | | vehicles | .user-mention} Tal | | | | ks | | | Ask Parking Services | to Parking | | | | Services | | | - Need 3 people. very hard to get | | | | people to the test track | - at least 3 | | | | people - 2 | | | | inside, one | | | | outside | | | | | +———+————————————–+——————–+
September 18th
+———+————————————–+——————–+ | Topic | Notes | Action Items | +=========+======================================+====================+ | CDR + | - Next Wednesday and Friday | \ | | Social | - CDR Technical Status | | | Respons | - 1-2 from continental + intel | | | ibility | - SAE reviewer | | | | - 2-3 from GM (mosty | | | | engineers) | | | | - Might be marketing | | | | background. | | | | - Explain explicitly the design | | | | space | | | | - Trading off engineering | | | | effort vs quality of user | | | | experience | | | | - PID = low engineering | | | | labour, low computation | | | | power, jerky experience. | | | | High in calibration | | | | - Dyn. MPC = high engineering | | | | labour, high computation | | | | power, best experience, high | | | | in calibration | | | | - Kyn. MPC = High engineering | | | | labour, medium computation | | | | power, best experience, low | | | | in calibration | | | | - High engineering labour, | | | | but it just works - end | | | | user requires no tuning | | | | - Don't need a physical | | | | car, and simulation will | | | | reflect the actual car | | | | accurately | | | | | | | | | | | | | | | | - When you have lots of systems | | | | competing for access, | | | | controllers will compete and | | | | cause problems | | | | - Create a "domain controller" - | | | | one big control strategy for | | | | everything. Main controller | | | | specifically requests certain | | | | things: "ABS: deal with L/R | | | | while AWD shift to front and | | | | back and everything you do must | | | | be through me so I can | | | | coordinate efforts | | | | - Research: 1 controller that does | | | | everything (not feasible in | | | | production as everything needs | | | | to be integrated) | | | | - Production: (~lvl2) too | | | | expensive. vendors create | | | | specific parts. SuperCruise is | | | | fully integrated | | | | - Controllers battle for | | | | ownership | | | | - Middleman domain controller | | | | who picks which controller | | | | to follow | | | | - Limit controller's | | | | knowledge of the environment | | | | | | | | | | | | | | | | - If we force all controllers to | | | | align with X, | | | | - AutoCzar can compile the MPC and | | | | keep implementation details and | | | | parameters hidden | | | | | | | | | | | | | | | | - Mohammed MPC: Refined exact | | | | position of hte car being in | | | | certain spot - smoother | | | | position. less for active safety | | | | but for smooth controls | | | | - Tire modeling doesn't work | | | | well with low speed and low | | | | slip → park assist | | | | | | | | | | | | | | | | - OEM: GM | | | | - Tier 1 Supplier: Bosch | | +———+————————————–+——————–+ | \ | \ | \ | +———+————————————–+——————–+ | \ | \ | \ | +———+————————————–+——————–+ | \ | \ | \ | +———+————————————–+——————–+
\
September 11th
+———+————————————–+——————–+ | Topic | Notes | Action Items | +=========+======================================+====================+ | Do we | - Not enough money for Rowan (can | \ | | have | use Derek's funds) | | | money | - Money for Components is fine, | | | | just need to make sure we | | | | actually redeem money | | | | - More money to pay grad students | | | | (Rowan) - do well in the | | | | competition | | +———+————————————–+——————–+ | Safety | - Need a decontamination zone - | \ | | Plan | leave stuff for a couple hours. | | | | - Latex gloves → Work Gloves for | | | | mechanical team. | | | | - Single use vs Multi use → is | | | | this time multiplexed? ie, | | | | single simultaneous or | | | | solo-assigned | | | | - Members are allowed to work | | | | alone as we are low risk | | | | - Derek gave a bunch of machines. | | | | Using one machine for | | | | RoadRunner, not using the | | | | others. 3 machines we can bring | | | | on (in Garage) if needed | | +———+————————————–+——————–+ | Practic | - Kevin will help review | \ | | e | presentations | | | Talk | - Mathworks Sun 20th, others Sept | | | | 26 | | +———+————————————–+——————–+ | In-vehi | - Need to validate simulation | \ | | cle | controls concepts work IRL | | | testing | - localization, mpc | | | | - Front load verification work so | | | | we know that sim represents real | | | | life | | | | - wipe down steering wheels and | | | | controls, windows open, shower | | | | curtains | | | | - Most testing can probably be | | | | done on campus. | | | | - For going to WRESTRC, only. two | | | | people: safety driver and | | | | pipeline operator | | +———+————————————–+——————–+
September 4th
+———+————————————–+——————–+
| Topic | Notes | Action Items |
+=========+======================================+====================+
| Car | - Everything Works. Can visit bay. | \ |
| Checkup | Small issue with air buildup in | |
| | coolant. Car battery is fine. | |
| | - Other teams: put a transparent | |
| | shower curtain in the car to | |
| | separate the front and the back. | |
+———+————————————–+——————–+
| Year | - Unfortunate Y3 issues. Lots of | \ |
| 3/4 | variation between schools about | |
| Checkup | what freedoms students have on | |
| | campus. | |
| | - Luckily for us, simulation is | |
| | great 
{.emoticon | |
| | .emoticon-laugh} | |
| | - In car testing should be | |
| | validation, not development | |
| | - Process of creating new safety | |
| | drivers is not possible. Is | |
| | there a replacement system?\ | |
| | - Kevin: No. Vehicles with | |
| | failure injection are not | |
| | available In Ca. | |
| | - Luckily we have 2 safety | |
| | drivers, Sam + Rowan | |
| | - Driver Training + US Border are | |
| | going to be difficult. | |
+———+————————————–+——————–+
| Year 4 | - Where are people and how | \ |
| | comfortable are they with the Y3 | |
| | Challenge? | |
| | - Pretty good with everything | |
| | but rerouting in Simulation. | |
| | Need to do controls tuning | |
| | in car | |
| | - Recruitment: 50 returning | |
| | people. Not really looking for | |
| | growth | |
+———+————————————–+——————–+
| Kevin | - Can help with vehicle plan - | \ |
| Resourc | shower curtain separating front | |
| es | and back. Windows open, HVAC | |
| | off. | |
| | - Kevin is pretty flexible as a | |
| | resource. Kevin is working as | |
| | tech lead on perception + | |
| | sensors. Lots of experience with | |
| | sensors | |
| | - Mentors: Competition is | |
| | reflective is reality. Make sure | |
| | that expectations are reasonable | |
| | and on an even playing field for | |
| | external factors. Relay that | |
| | back to the organizing facility. | |
| | Need information from GM / SAE | |
| | - Contact to suppliers | |
| | (Continental) | |
| | - Competition Roadmap | |
| | - Mathworks Competition | |
+———+————————————–+——————–+
| Localiz | - Use a GPS at regular speed | \ |
| ation | (~20km/h) to see the | |
| | coefficient for wheel size. then | |
| | use this for correction. | |
| | - Measure wheel center to road. | |
| | Jack the car up and measure | |
| | wheel diameter. Use 3/4 of that. | |
| | As you go faster, wheels get | |
| | bigger | |
| | - Park assist use wheel + steering | |
| | angle | |
| | - Diverges. Have a vehicle model. | |
| | Gets pretty bad after a couple | |
| | seconds with just odometry. Need | |
| | visual corrections | |
| | - Kevin approves of Occupancy | |
| | Grid. | |
| | - Encoders are pretty accurate if | |
| | you're at >4-5km/h | |
| | - Really slow isn't great | |
| | - Kevin review our static Y3 | |
| | challenges | |
+———+————————————–+——————–+
| Recomme | Do sanity checking of simulation ↔ | \ |
| ndation | car BEFORE it snows. | |
+———+————————————–+——————–+
Document generated by Confluence on Nov 19, 2021 12:21