Skip to content

Commit df897ac

Browse files
kiwi-sherbetkiwi-sherbet
authored
revised minor layout
1 parent c17f5e9 commit df897ac

File tree

1 file changed

+117
-121
lines changed

1 file changed

+117
-121
lines changed

‎index.markdown‎

Lines changed: 117 additions & 121 deletions
Original file line numberDiff line numberDiff line change
@@ -142,29 +142,29 @@ highlight {
142142
<center><p><span style="font-size:20px;"></span></p></center>
143143

144144
<table border="0" cellspacing="10" cellpadding="0" align="center">
145-
<tbody>
146-
<tr>
147-
<td align="center" valign="middle">
145+
<tbody>
146+
<tr>
147+
<td align="center" valign="middle">
148148
<video muted autoplay loop width="798">
149149
<source src="./src/video/header.mp4" type="video/mp4">
150150
</video>
151151
</td>
152152
</tr>
153153
</tbody>
154154
</table>
155-
<table border="0" cellspacing="10" cellpadding="0" align="center">
156-
<tr>
157-
<td>
158-
<p align="justify" width="20%">
159-
We tackle the problem of developing humanoid loco-manipulation skills with deep imitation learning. The challenge of collecting human demonstrations for humanoids, in conjunction with the difficulty of policy training under a high degree of freedom, presents substantial challenges.
160-
We introduce <b>TRILL</b>, a data-efficient framework for learning humanoid loco-manipulation policies from human demonstrations.
161-
In this framework, we collect human demonstration data through an intuitive Virtual Reality (VR) interface.
162-
We employ the whole-body control formulation to transform task-space commands from human operators into the robot's joint-torque actuation while stabilizing its dynamics.
163-
By employing high-level action abstractions tailored for humanoid robots, our method can efficiently learn complex loco-manipulation skills.
164-
We demonstrate the effectiveness of TRILL in simulation and on a real-world robot for performing various types of tasks.
165-
</p>
166-
</td>
167-
</tr>
155+
<table align=center width=800px>
156+
<tr>
157+
<td>
158+
<p align="justify" width="20%">
159+
We tackle the problem of developing humanoid loco-manipulation skills with deep imitation learning. The challenge of collecting human demonstrations for humanoids, in conjunction with the difficulty of policy training under a high degree of freedom, presents substantial challenges.
160+
We introduce <b>TRILL</b>, a data-efficient framework for learning humanoid loco-manipulation policies from human demonstrations.
161+
In this framework, we collect human demonstration data through an intuitive Virtual Reality (VR) interface.
162+
We employ the whole-body control formulation to transform task-space commands from human operators into the robot's joint-torque actuation while stabilizing its dynamics.
163+
By employing high-level action abstractions tailored for humanoid robots, our method can efficiently learn complex loco-manipulation skills.
164+
We demonstrate the effectiveness of TRILL in simulation and on a real-world robot for performing various types of tasks.
165+
</p>
166+
</td>
167+
</tr>
168168
</table>
169169

170170
<hr>
@@ -173,148 +173,144 @@ highlight {
173173
<table border="0" cellspacing="10" cellpadding="0" align="center">
174174
<tbody>
175175
<tr>
176-
<td align="center" valign="middle">
177-
<a href="./src/figure/approach.png"><img src="./src/figure/approach.png" style="width:100%;"> </a>
176+
<td align="center" valign="middle">
177+
<a href="./src/figure/approach.png"><img src="./src/figure/approach.png" style="width:100%;"> </a>
178178
</td>
179179
</tr>
180180
</tbody>
181181
</table>
182182
<table align=center width=800px>
183-
<tr>
184-
<td>
185-
<p align="justify" width="20%">
186-
TRILL addresses the challenge of learning humanoid loco-manipulation.
187-
We introduce a learning framework that facilitates teleoperated demonstrations with task-space commands provided by a human demonstrator.
188-
The trained policies leverage human complexity and adaptability in decision-making to generate these commands.
189-
The robot control interface then executes these target commands through joint-torque actuation, complying with robot dynamics.
190-
This synergistic combination of imitation learning and whole-body control enables successful method implementation in both simulated and real-world environments.
191-
</p>
192-
</td>
193-
</tr>
183+
<tr>
184+
<td>
185+
<p align="justify" width="20%">
186+
TRILL addresses the challenge of learning humanoid loco-manipulation.
187+
We introduce a learning framework that facilitates teleoperated demonstrations with task-space commands provided by a human demonstrator.
188+
The trained policies leverage human complexity and adaptability in decision-making to generate these commands.
189+
The robot control interface then executes these target commands through joint-torque actuation, complying with robot dynamics.
190+
This synergistic combination of imitation learning and whole-body control enables successful method implementation in both simulated and real-world environments.
191+
</p>
192+
</td>
193+
</tr>
194194
</table>
195195

196196
<hr>
197197

198198
<h1 align="center">Hierarchical Loco-manipulation Pipeline</h1>
199199
<table border="0" cellspacing="10" cellpadding="0" align="center">
200-
<tbody>
201-
<tr>
202-
<td align="center" valign="middle">
203-
<a href="./src/figure/pipeline.png">
204-
<img src="./src/figure/pipeline.png" style="width:100%;">
205-
</a>
206-
</td>
207-
</tr>
208-
</tbody>
200+
<tbody>
201+
<tr>
202+
<td align="center" valign="middle">
203+
<a href="./src/figure/pipeline.png">
204+
<img src="./src/figure/pipeline.png" style="width:100%;">
205+
</a>
206+
</td>
207+
</tr>
208+
</tbody>
209209
</table>
210-
211210
<table align=center width=800px>
212-
<tr>
213-
<td>
214-
<p align="justify" width="20%">
215-
The trained policies generate the target task-space command at 20Hz from the onboard stereo camera observation and the robot's proprioceptive feedback.
216-
The robot control interface realizes the task-space commands and computes the desired joint torques at 100Hz and sends them to the humanoid robot for actuation.
217-
More implementation details can be found in <a href="https://github.com/UT-Austin-RPL/TRILL/blob/main/docs/Implementation-Details.md">this page</a>.
218-
</p>
219-
</td>
220-
</tr>
211+
<tr>
212+
<td>
213+
<p align="justify" width="20%">
214+
The trained policies generate the target task-space command at 20Hz from the onboard stereo camera observation and the robot's proprioceptive feedback.
215+
The robot control interface realizes the task-space commands and computes the desired joint torques at 100Hz and sends them to the humanoid robot for actuation.
216+
More implementation details can be found in <a href="https://github.com/UT-Austin-RPL/TRILL/blob/main/docs/Implementation-Details.md">this page</a>.
217+
</p>
218+
</td>
219+
</tr>
221220
</table>
222221

223222
<hr>
224223

225224
<h1 align="center">Real-Robot Teleoperation</h1>
226225
<table align=center width=800px>
227-
<tr>
228-
<td>
229-
<p align="justify" width="20%">
230-
We design an intuitive VR teleoperation system, which reduces the cognitive and physical burdens for human operators to provide task demonstration.
231-
As a result, our teleoperation approach can produce high-quality demonstration data while maintaining safe robot operation.
232-
</p>
233-
</td>
234-
</tr>
226+
<tr>
227+
<td>
228+
<p align="justify" width="20%">
229+
We design an intuitive VR teleoperation system, which reduces the cognitive and physical burdens for human operators to provide task demonstration.
230+
As a result, our teleoperation approach can produce high-quality demonstration data while maintaining safe robot operation.
231+
</p>
232+
</td>
233+
</tr>
235234
</table>
236235
<table border="0" cellspacing="10" cellpadding="0" align="center">
237-
<tbody>
238-
<tr>
239-
<td align="center" valign="middle">
240-
<video controls width="798">
241-
<source src="./src/video/demo_ramen.mp4" type="video/mp4">
242-
</video>
243-
</td>
244-
</tr>
245-
</tbody>
236+
<tbody>
237+
<tr>
238+
<td align="center" valign="middle">
239+
<video controls width="798">
240+
<source src="./src/video/demo_ramen.mp4" type="video/mp4">
241+
</video>
242+
</td>
243+
</tr>
244+
</tbody>
246245
</table>
247-
248246
<table align=center width=800px>
249-
<tr>
250-
<td>
251-
<p align="justify" style="text-align: right;" width="20%">
252-
Music: <a href="https://soundcloud.com/bergscloud/happy">Happy</a> by <a href="https://soundcloud.com/bergscloud">Luke Bergs</a>
253-
</p>
254-
</td>
255-
</tr>
247+
<tr>
248+
<td>
249+
<p align="justify" style="text-align: right;" width="20%">
250+
Music: <a href="https://soundcloud.com/bergscloud/happy">Happy</a> by <a href="https://soundcloud.com/bergscloud">Luke Bergs</a>
251+
</p>
252+
</td>
253+
</tr>
256254
</table>
257255

258256
<hr>
259257

260258
<h1 align="center">Real-Robot Deployment</h1>
261259
<table align=center width=800px>
262-
<tr>
263-
<td>
264-
<p align="justify" width="20%">
265-
We demonstrate the application of TRILL on the real robot, deploying visuomotor policies trained for dexterous manipulation tasks.
266-
During evaluation, the robot performed each task 10 times in a row without rebooting and succeeded in 8 out of 10 trials in the <i>Tool pick-and-place</i> task and 9 out of 10 trials in the <i>Removing the spray cap</i> task, respectively.
267-
</p>
268-
</td>
269-
</tr>
260+
<tr>
261+
<td>
262+
<p align="justify" width="20%">
263+
We demonstrate the application of TRILL on the real robot, deploying visuomotor policies trained for dexterous manipulation tasks.
264+
During evaluation, the robot performed each task 10 times in a row without rebooting and succeeded in 8 out of 10 trials in the <i>Tool pick-and-place</i> task and 9 out of 10 trials in the <i>Removing the spray cap</i> task, respectively.
265+
</p>
266+
</td>
267+
</tr>
270268
</table>
271-
272269
<table border="0" cellspacing="10" cellpadding="0" align="center">
273-
<tbody>
274-
<tr>
275-
<td align="center" valign="middle">
276-
<video muted controls width="394">
277-
<source src="./src/video/deploy_box.mp4" type="video/mp4">
278-
</video>
279-
</td>
280-
<td align="center" valign="middle">
281-
<video muted controls width="394">
282-
<source src="./src/video/deploy_cap.mp4" type="video/mp4">
283-
</video>
284-
</td>
285-
</tr>
286-
</tbody>
270+
<tbody>
271+
<tr>
272+
<td align="center" valign="middle">
273+
<video muted controls width="394">
274+
<source src="./src/video/deploy_box.mp4" type="video/mp4">
275+
</video>
276+
</td>
277+
<td align="center" valign="middle">
278+
<video muted controls width="394">
279+
<source src="./src/video/deploy_cap.mp4" type="video/mp4">
280+
</video>
281+
</td>
282+
</tr>
283+
</tbody>
287284
</table>
288285

289286
<hr>
290287

291288
<h1 align="center">Simulation Evaluation</h1>
292289
<table align=center width=800px>
293-
<tr>
294-
<td>
295-
<p align="justify" width="20%">
296-
We design two realistic simulation environments and evaluate the robot’s ability to successfully perform subtasks involving free-space locomotion, manipulation, and loco-manipulation.
297-
TRILL, a framework tailored to train humanoid robots, achieves success rates of 96% for free-space locomotion tasks, 80% for manipulation tasks, and 92% for loco-manipulation tasks.
298-
</p>
299-
</td>
300-
</tr>
290+
<tr>
291+
<td>
292+
<p align="justify" width="20%">
293+
We design two realistic simulation environments and evaluate the robot’s ability to successfully perform subtasks involving free-space locomotion, manipulation, and loco-manipulation.
294+
TRILL, a framework tailored to train humanoid robots, achieves success rates of 96% for free-space locomotion tasks, 80% for manipulation tasks, and 92% for loco-manipulation tasks.
295+
</p>
296+
</td>
297+
</tr>
301298
</table>
302-
303299
<table border="0" cellspacing="10" cellpadding="0" align="center">
304-
<tbody>
305-
<tr>
306-
<td align="center" valign="middle">
307-
<video muted controls width="394">
308-
<source src="./src/video/deploy_door.mp4" type="video/mp4">
309-
</video>
310-
</td>
311-
<td align="center" valign="middle">
312-
<video muted controls width="394">
313-
<source src="./src/video/deploy_workbench.mp4" type="video/mp4">
314-
</video>
315-
</td>
316-
</tr>
317-
</tbody>
300+
<tbody>
301+
<tr>
302+
<td align="center" valign="middle">
303+
<video muted controls width="394">
304+
<source src="./src/video/deploy_door.mp4" type="video/mp4">
305+
</video>
306+
</td>
307+
<td align="center" valign="middle">
308+
<video muted controls width="394">
309+
<source src="./src/video/deploy_workbench.mp4" type="video/mp4">
310+
</video>
311+
</td>
312+
</tr>
313+
</tbody>
318314
</table>
319315

320316
<hr>
@@ -326,9 +322,9 @@ highlight {
326322
<pre><code style="display:block; overflow-x: auto">
327323
@inproceedings{seo2023trill,
328324
title={Deep Imitation Learning for Humanoid Loco-manipulation
329-
through Human Teleoperation},
325+
through Human Teleoperation},
330326
author={Seo, Mingyo and Han, Steve and Sim, Kyutae and Bang, Seung Hyeon
331-
and Gonzalez, Carlos and Sentis, Luis and Zhu, Yuke},
327+
and Gonzalez, Carlos and Sentis, Luis and Zhu, Yuke},
332328
booktitle={IEEE-RAS International Conference on Humanoid Robots (Humanoids)},
333329
year={2023}
334330
}

0 commit comments

Comments
 (0)