What Happened at the DARPA Robotics Challenge Finals

Atkeson, Christopher G.; Benzun, P. W. Babu; Banerjee, Nandan; Berenson, Dmitry; Bove, Christoper P.; Cui, Xiongyi; DeDonato, Mathew; Du, Ruixiang; Feng, Siyuan; Franklin, Perry; Gennert, Michael A.; Graff, Joshua P.; He, Peng; Jaeger, Aaron M.; Kim, Joohyung; Knoedler, Kevin; Li, Lening; Liu, Chenggang; Long, Xin; Padır, Taşkın; Polido, Felipe; Tighe, G. G.; Xinjilefu, X

doi:10.1007/978-3-319-74666-1_17

Cited by 52 publications

(42 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1) Generate off-line N valid sample limb configurations q k i , 0 ≤ i < N (We choose N = 10 4 ); 2) Using the end-effector positions p(q k i ) as indices, store each sample in an octree data structure; 3) At runtime, when contact creation is required, intersect the octree and the environment 5 to retrieve the list of samples S ⊂ C ǫ Contact close to contact (Figure 5 (b) and (c)); 4) Use a user-defined heuristic h to sort S; 5) If S is empty, stop (failure). Else select the first configuration of S. Project it onto contact using inverse kinematics.…”

Section: Contact Generatormentioning

confidence: 99%

“…For this reason previous contributions having demonstrated acyclic contact locomotion on a real robot are too computationally expensive [4]. As a result at the DARPA Robotics Challenge, the participants stated that except for egress, the robots did not use multi contact strategies: they relied on unsafe bipedal walking to climb stairs, instead of using the provided handrails to facilitate the motion [5].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

An Efficient Acyclic Contact Planner for Multiped Robots

et al. 2018

View full text Add to dashboard Cite

We present a contact planner for complex legged locomotion tasks: standing up, climbing stairs using a handrail, crossing rubble and getting out of a car. The need for such a planner was shown at the Darpa Robotics Challenge, where such behaviors could not be demonstrated (except for egress). Current planners suffer from their prohibitive algorithmic complexity, because they deploy a tree of robot configurations projected in contact with the environment. We tackle this issue by introducing a reduction property: the reachability condition. This condition defines a geometric approximation of the contact manifold, which is of low dimension, presents a Cartesian topology, and can be efficiently sampled and explored. The hard contact planning problem can then be decomposed into two sub-problems: first, we plan a path for the root without considering the whole-body configuration, using a sampling-based algorithm; then, we generate a discrete sequence of whole-body configurations in static equilibrium along this path, using a deterministic contact-selection algorithm. The reduction breaks the algorithm complexity encountered in previous works, resulting in the first interactive implementation of a contact planner (open source). While no contact planner has yet been proposed with theoretical completeness, we empirically show the interest of our framework: in a few seconds, with high success rates, we generate complex contact plans for various scenarios and two robots, HRP-2 and HyQ. These plans are validated either in dynamic simulations, or on the real HRP-2 robot.

show abstract

Section: Contact Generatormentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

An Efficient Acyclic Contact Planner for Multiped Robots

et al. 2018

View full text Add to dashboard Cite

show abstract

“…A report on the performance of teams in the DARPA Robotics Challenge (Atkeson et al, 2015) stated that one of the major causes of failure was robots' inability to handle small variations in their tasks, a problem that arose when tests were no longer performed in a laboratory setting.…”

Section: Research Trends and Directionsmentioning

confidence: 99%

Autonomous robots for harsh environments: a holistic overview of current solutions and ongoing challenges

Wong

Yang

Yan

et al. 2018

Systems Science & Control Engineering

View full text Add to dashboard Cite

This paper presents a holistic overview of robotics and autonomous systems developed for applications in harsh environments where interactions with robots are difficult to model due to reasons such as dynamics, uncertainty, complexity and unpredictability. Robots of different characteristics and designs are required to perform different tasks within these environments. A number of classes of robots emerge as particularly favoured by the research community for solving challenging problems in difficult environments. Key examples of these deployments are studied with an analysis on how high-level autonomy is a key issue that must be addressed. The surveyed work suggest that the lack of observable autonomy in many systems are a consequence of both the complexity of the problem and the lack of proven reliability of autonomous solutions for applications that demand high success rates. This paper provides a broad and general overview of autonomous robots deployed in harsh environments and the commonalities of the challenges and existing solutions across application areas, including oil and gas inspection, space exploration, deep-sea operations and search and rescue. ARTICLE HISTORY

show abstract

“…Despite DNN's massive computational capabilities, there are two aspects that temper these accomplishments: first, the number of training samples required to reach acceptable performances is huge -tens or hundreds of millions; second, these architectures show a limited ability to generalize to new tasks/settings that were not encountered during training. These limitations become largely evident in motor control as shown by the clumsy behaviour of humanoid robots in the DARPA robotic challenge [4] .…”

Section: Introductionmentioning

confidence: 99%

The neural and cognitive architecture for learning from a small sample

Cortese

Martino

Kawato

2019

Current Opinion in Neurobiology

View full text Add to dashboard Cite

Artificial intelligence algorithms are capable of fantastic exploits, yet they are still grossly inefficient compared with the brain's ability to learn from few exemplars or solve problems that have not been explicitly defined. What is the secret that the evolution of human intelligence has unlocked? Generalization is one answer, but there is more to it. The brain does not directly solve difficult problems, it is able to recast them into new and more tractable problems. Here we propose a model whereby higher cognitive functions profoundly interact with reinforcement learning to drastically reduce the degrees of freedom of the search space, simplifying complex problems and fostering more efficient learning.

show abstract

What Happened at the DARPA Robotics Challenge Finals

Cited by 52 publications

References 4 publications

An Efficient Acyclic Contact Planner for Multiped Robots

An Efficient Acyclic Contact Planner for Multiped Robots

Autonomous robots for harsh environments: a holistic overview of current solutions and ongoing challenges

The neural and cognitive architecture for learning from a small sample

Contact Info

Product

Resources

About