CS 685
Project Guidelines

Here are some proposals for the class projects. Your best bet is to pick a topic which explores in greater detail some topics covered in class (hence involes additional reading) and implement the chosen algorithm and demostrate it in action.

In case you do not plan to select one of the suggested projects, your project description is due in class last week on October. You can also schedule a meeting with me to discuss the project, if you are not sure what to do or need some help specifying details. The requirement of the project is that you will have to study, read some advanced material (either papers or advanced chapters from the books) not covered in the class and have a very clear idea about the outcome of your project.

In the project proposal description you should specify the following components:

  • Project Guidelines: What is the goal of the project.
  • Data: What type of data, algorithms will you use.
  • Tasks: What are the indiviudal stages of the project, i.e. tasks you plan to accomplish.

    Additional project ideas:

  • Learning how to explore indoors environments and obstacle avoidance in the simulator http://gibsonenv.stanford.edu/
  • Improve the effectiveness of model based strategies vs learned stategies using CARLA simulator (see the paper)
  • Volumetric reconstruction of Active Vision Dataset http://cs.unc.edu/~ammirato/active_vision_dataset_website/index.html using this library http://www.open3d.org/
  • Detection of anomalies in automated driving videos using some data - https://deepdrive.berkeley.edu
  • Training your own object detector for a household object of your choice - https://github.com/tensorflow/models/tree/master/research/object_detection

    • Previous Projects

  • Simulating Pathfinding in Unity3D or PyBullet using RRT
  • Particle filter localization robots
  • Grocery Products Recognition
  • NBNL based Sports Scene Classification Using pre-trained CNN Features
  • Path planning in dynamic environments
  • Hidden Markov Model for Continuous Ultrasound-based Gesture Recognition

    • Recognition, Localization and Mapping

  • ICP algorithm (Iterative Closest Point Algorithm). Implement the motion estimation using the basic Iterative Closest Point (ICP) Algorithm, which aligns the two 3D scans. Evaluate its performance and choices of different parameters on several datasets or range data, or combined range and visual data Input: sequence of range scans and imaeges acquired by Kinect sensor along with synchronized sequences of color images. Output: computed motions between consecutive frames and visualization of the aligned cloud points.
  • Particle filter. In this project you implement complete particle filter. Input: You will be given the representation of world along with the sensor readings to be used by the filter and some code snippets. Output: You will have to implement the remaining portions of the code to have the filter working properly.
  • Vision Based Localization using scale invariant features reading
  • Recognizing groceries using vision reading
  • Particle Filter for Localization using Range data (simulation) reading1 , reading2
  • Extended Kalman filter In this project you implement complete Extended kalman filter for estimation of motion from features tracks (or other sensing modality). Input/Output will depend of the chosen sensor model.
  • Visual Odometry. Given a stream of images, compute the trajectory of the mobile platform
  • 3D object detection. Implement car detection algorithm which can effectively use 3D information available from laser range data along with the images
  • Independently moving object detection. Given a video sequence, use feature tracking techniques to implement an algorithm for independent motion detection
  • Gesture Recognition. Using skeletal tracking on Kinect SDK, propose and implement simple human gesture recognition system for human robot interface

    Path planning, Sensor Based Planning, Sampling based planning

  • Implement exploration strategy for a robot to explore 4th floor of the Engineering Building (implement in simulation)
  • Pick two different path planning algorithms on a grid. (Best first search, A*), try them on different environments, generalize to higher dimension or multiple robots. Path planning algorithms have to be demonstrated on some nontrivial environments.
  • Sensor Based Motion Planning: The Hierarchical Generalized Voronoi Graph , WAFR 96., Incremental Construction of Generalized Voronoi Graph, ICRA 95
  • Sensor based path planning in the dynamically changing or partially unknown environments. A. Stentz D* - Dynamic A* algorithm (.pdf) and its extensions. Experiment with different versions of the cost function (time traveled, danger), multiple robots
  • Extensions of RRT's to moving objects and/or feedback motion strategies, Steven LaValle; Read Chap 7., Chap 8. of the book. L

    Multi-robot problems, Coverage problems, tasks, architectures

  • Combine reactive and path planning strategies to achieve the best possible coverage of a particular area.
  • Design decentralized conrol strategies for formation flight
  • Formation Control using Graph Based Techniques
  • Design reactive behaviors based on potential field strategies to acomplish set cooperative behaviors for mutiple mobile robot agents. Experiment with different sensing and communication strategies. Experiment with different arbitration strategies.
    (see paper by Craig Reynolds on Steering Behaviors for Autonomous characters.)

    Inverse Kinematics

    Inverse Kinematics Survey paper