CVPR23 E2EAD | Phil Duan, Invited Talk

Phil Duan's talk: Building Vision Foundation Models for Autonomous Driving at CVPR23 E2EAD Workshop, hosted by @OpenDriveLab (opendrivelab.com)
"Correction: 04:18 That should be 40 deg-of-freedom body control instead of 200"
AD Challenge: opendrivelab.com/AD23Challeng...
CVPR23 E2EAD Workshop: opendrivelab.com/e2ead/cvpr23...

[@philforrence]

Nice!

[@pervezbhan1708]

There are several algorithms and techniques used for predictions and planning in autonomous vehicles. Here are some commonly employed methods:
Predictions Algorithms:
a. Kalman Filters and Extended Kalman Filters: These recursive estimation algorithms are widely used for sensor fusion and tracking the motion of objects based on noisy sensor measurements.
b. Particle Filters: They are used for state estimation and tracking, particularly when dealing with non-linear and non-Gaussian systems.
c. Hidden Markov Models (HMM): HMMs are probabilistic models used for predicting the future behavior of objects by considering their current states and previous observations.
d. Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM): These deep learning architectures can be used to learn temporal patterns from historical sensor data and predict future trajectories.
e. Social Force Models: These models simulate the interactions between pedestrians and vehicles by considering social behaviors and physical forces.
Planning Algorithms:
a. A* (A-Star) Algorithm: A* is a popular graph search algorithm used for finding the shortest path between two points in a graph representation of the environment.
b. RRT (Rapidly-Exploring Random Trees): RRT is a sampling-based algorithm that efficiently explores the configuration space of a vehicle and generates feasible paths by incrementally growing a tree.
c. Model Predictive Control (MPC): MPC is a control algorithm that plans optimal trajectories by predicting the system's future behavior and iteratively optimizing control inputs to minimize a defined cost function.
d. Dynamic Programming: Dynamic programming algorithms, such as Value Iteration and Policy Iteration, can be used for planning optimal paths by solving a sequence of subproblems.
e. Reinforcement Learning: Reinforcement learning algorithms, such as Deep Q-Networks (DQN) and Proximal Policy Optimization (PPO), can learn to make planning decisions through interactions with the environment and receiving rewards or penalties.
It's worth noting that autonomous driving systems often employ a combination of these algorithms and techniques, and the specific choices may vary depending on the complexity of the environment, the level of autonomy, and the available sensor suite. Additionally, ongoing research and advancements in machine learning, optimization, and robotics continue to contribute to the development of more sophisticated and efficient predictions and planning algorithms for autonomous vehicles.

[@Supreme_Lobster]

thanks this is a great pointer

[@OpenDriveLab]

Correction: 04:18 That should be 40 deg-of-freedom body control instead of 200.

[@danypell2517]

ooooo!

[@user-zt7kj4um1l]

악

[@tylerbelyeu846]

First? 😂