To view the documents in this web-page you'll require Adobe Acrobat Reader, Open Office, VLC media player

Updates on Works done on the LittleDog Robot
Subhrajit Bhattacharya
subhrabh_AT_seas.upenn.edu

------------------------------------------------------------
Latest version of the report: report.pdf
Latest version of MATLAB code: code.zip
Presentation: presentation_for_website.odp (Additional files for presentation: All the movie files in this web-page + Movie_1.wmv + Movie_2.wmv)
------------------------------------------------------------

Movies of Simulations and Experiments:
(Apologies for the low quality of the movies caused by the very minimal compression software used for compressing the movie files)

Local Planning
Hard persuasion strategy (Based on algorithm of Section 8.1 of the report)
Flat terrain, Straight-line trajectory
No feedback correction:
Simulation: DogData_Flat_Hard_sim1.mpg (1.1 mb)
Run on LittleDog robot: DodData_Flat_Hard_movie1.mpg (3.3 mb)
Notes:
We use the hard persuasion for planning motion on a flat terrain and along a simple straight-line trajectory without any feedback correction.
As expected from the hard persuasion, the robot performs exact tracking of the given trajectory. Hence the height of its center of mass remains fixed from the ground, since that's how the trajectory was chosen.

Local Planning
Soft persuasion strategy (Based on algorithm of Section 8.2 of the report)
Flat terrain, Straight-line trajectory
No feedback correction:
Case - I:
Simulation: DogData_Flat_1_nofeedback_old_sim1.mpg (0.8 mb)
Run on LittleDog robot: DogData_Flat_1_nofeedback_old_movie1.mpg (2.5 mb)
Case - II:
Simulation: DogData_Flat_7_noFeedback_sim1.wmv (4.5 mb)
Run on LittleDog robot: DogData_Flat_7_noFeedback_movie1_2x.mpg (2.6 mb - Speed: 2x)
Notes:
We use the soft persuasion for planning motion on a flat terrain and along a simple straight-line trajectory without any feedback correction.
As expected from the soft persuasion, the center of mass of the robot deviates away from the trajectory at times.
The primary difference between case-I and case-II is that in the former case we use small lengths for the robot's step size, while in the later case we use longer steps for the robot's foot placement.

Local Planning
Demonstration of Feedback Correction System (As described in Section 8.4 of the report)
Soft persuasion strategy (Based on algorithm of Section 8.2 of the report)
Flat terrain:
Comparison between desired robot state and actual robot state - without feedback: DogData_Flat_4_noFeedBack_ComparingMovie.wmv (5.6 mb)
Comparison between desired robot state and actual robot state - with feedback: DogData_Flat_4_yesFeedBack_ComparingMovie.wmv (5.6 mb)
Notes:
These videos essentially demostrate the working of a correction system that we have implemented to make the robot follow a pre-calculated sequence of steps using the feedback obtained from the MOCAP system. In either of the movies there are two skeletons of the LittleDog robot. One of them represents the desired robot state, while the other represents the experimental robot state as read from the MOCAP system. The experimental robot can be identified by a faint red circle at its center and by its flickering nature (which is caused by the noise in the readings obtained from the MOCAP).
It can be noted that although in either of the cases the error at the beginning are almost the same (compare the very first frames of the movies), in the case without a feedback correction the error persists and tends to increase, while in the case with a feedback correction system the error gets reduced to a low value very quickly. Moreover it can be noted that in case of introduction of further errors during the run-time, the feedback correction system actually makes necessary corrections so that the error is reduced withing a few steps.

Local Planning
Soft persuasion strategy (Based on algorithm of Section 8.2 of the report)
Feedback correction (As described in Section 8.4 of the report)
Rough terrain (terrain-B):
Simulation: DogData_TerrainB_5_sim1_2x.mpg (3.6 mb)
Run on LittleDog robot: DogData_TerrainB_5_movie1_4x.mpg (3.6 mb - Speed: 4x)
Notes:
The simulation shows walking along a given trajectory over 'terrain-B' using the algorithm of section 8.2.
We make the robot follow the states as computed in the simulation. We make use of a quick feedback correction system (as described in Section 8.4 of the report) to make any required correction for the deviation of the robot from the desired states. As quite evident from the movie, there are several flaws in our present feedback system. We are in the process of its development, and hope to come up with better results in the recent future.